Ontemporary Athematics: Differential Algebra, Complex Analysis and Orthogonal Polynomials

CONTEMPORARY
MATHEMATICS
509
Differential Algebra,
Complex Analysis
and Orthogonal Polynomials
Jairo Charris Seminar 2007–2008
Escuela de Matemáticas
Universidad Sergio Arboleda
Bogotá, Colombia
Primitivo B. Acosta-Humánez
Francisco Marcellán
Editors
American Mathematical Society

Instituto de Matemáticas y sus Aplicaciones

CONTEMPORARY
MATHEMATICS
509
Differential Algebra,
Complex Analysis
and Orthogonal Polynomials
Jairo Charris Seminar 2007–2008
Escuela de Matemáticas
Universidad Sergio Arboleda
Bogotá, Colombia
Primitivo B. Acosta-Humánez
Francisco Marcellán
Editors

Instituto de Matemáticas y sus Aplicaciones

Providence, Rhode Island
Editorial Board
Dennis DeTurck, managing editor
George Andrews Abel Klein Martin J. Strauss
2000 Mathematics Subject Classification. Primary 05C20, 12H05, 14E20, 14L99, 14M15,
20C20, 30C45, 33C50, 33D45, 34A26, 34C07, 34C08, 34M15, 35C05, 41A60, 42C05,
46E25, 53C15, 54C40.
Library of Congress Cataloging-in-Publication Data

Jairo Charris Seminar (2007–2008 : Universidad Sergio Arboleda)
Differential algebra, complex analysis and orthogonal polynomials : Jairo Charris Seminar,
2007–2008, Escuela de Matemáticas, Universidad Sergio Arboleda, Bogotá, Colombia / Primi-
tivo B. Acosta-Humánez, Francisco Marcellán, editors.
p. cm. — (Contemporary mathematics ; v. 509)
Includes bibliographical references.
ISBN 978-0-8218-4886-9 (alk. paper)
1. Differentiable dynamical systems—Congresses. 2. Orthogonal polynomials—Congresses.
I. Acosta-Humánez, Primitivo B. II. Marcellán, Francisco. III. Title.
QA614.8.J35 2010
515.39—dc22
2009043176
Copying and reprinting. Material in this book may be reproduced by any means for edu-
cational and scientific purposes without fee or permission with the exception of reproduction by
services that collect fees for delivery of documents and provided that the customary acknowledg-
ment of the source is given. This consent does not extend to other kinds of copying for general
distribution, for advertising or promotional purposes, or for resale. Requests for permission for
commercial use of material should be addressed to the Acquisitions Department, American Math-
ematical Society, 201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can
also be made by e-mail to reprint-permission@ams.org.
Excluded from these provisions is material in articles for which the author holds copyright. In
such cases, requests for permission to use or reprint should be addressed directly to the author(s).
(Copyright ownership is indicated in the notice in the lower right-hand corner of the first page of
each article.)

c 2010 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Copyright of individual articles may revert to the public domain 28 years
after publication. Contact the AMS for copyright status of individual articles.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10 9 8 7 6 5 4 3 2 1 15 14 13 12 11 10
In memory of Jairo Antonio Charris Castañeda, 1939-2003.
This page intentionally left blank
Contents
Preface vii
Differential Galois theory of algebraic Lie-Vessiot systems
D. Blázquez-Sanz and J. J. Morales-Ruiz 1
Recent trends on two variable orthogonal polynomials
L. Fernández, F. Marcellán, T. E. Pérez, and M. A. Piñar 59
On the integrability of the Riccati equation
C. A. Gomez S. 87
Two discrete systems of q-orthogonal polynomials
M. E. H. Ismail 95
Like-hyperbolic Bloch-Bergman classes
awrynowicz, L. F. Reséndis O., and L. M. Tovar S.
J. L 103
Some words about the application of Tchebycheff systems to Weak Hilbert’s
16th Problem
J. Tomás Lázaro 119
From the index of a differential operator to the Milnor number of a singularity
D. Mond 129
Integrability of dynamical systems through differential Galois theory: A
practical guide
J. J. Morales-Ruiz and J.-P. Ramis 143
Tournaments and parabolic almost complex structures on flag manifolds
M. Paredes and S. Pinzón 221
v
Preface
This volume contains articles by nine invited speakers in the Jairo Charris
Seminar in Algebra and Analysis held at the Universidad Sergio Arboleda from
August 1 to August 3, 2007 and from July 31 to August 2, 2008 to honor Professor
Jairo Charris Castañeda, who died in 2003 as a consequence of a serious illness.
Professor Charris was one of the most relevant Colombian mathematicians in
the last third of the past century. He was born in Cienaga, Department of Mag-
dalena, Colombia, in 1939. He obtained a Master’s degree in Chemical Engineering
in 1962 and in Mathematics in 1967, both from Universidad Nacional de Colom-
bia. In 1969, he received a Master’s degree in Mathematics from the University of
Chicago under the advisoring of Professor R. Narasimhan. From 1981, and under
the direction of Professor M. E. H. Ismail, he worked in the Department of Mathe-
matics at Arizona State University in the field of Pollaczek orthogonal polynomials.
There, in 1984 he defended his PhD Thesis. He came back to Colombia to con-
tinue his scientific career as a professor at Universidad Nacional at Bogotá until his
retirement in 1998. After retiring, he collaborated with Universidad Nacional and
Universidad Sergio Arboleda both in teaching and research activities.
The scientific fields where Professor Charris focused his activity were the theory
of orthogonal polynomials (in particular, his contributions on sieved polynomials
had a strong impact in the mathematical community interested in the subject),
Complex Analysis, Compactification Theory, and Group Algebra. He authored 32
publications and three monographs. His scientific leadership is very remarkable,
with 20 graduate and postgraduate students who defended Master’s and PhD The-
ses under his direction from 1976 to 2001.
The articles contained in this volume cover a wide range of topics in the theory
of Integrable Dynamical Systems based on different approaches such as Differential
Galois Theory and Lie Groups, as well as some recent developments in the theory
of multivariable and q-orthogonal polynomials, Weak Hilbert’s 16th Problem, Sin-
gularity Theory, Tournaments in flag manifolds, and spaces of bounded analytic
functions on the unit circle. The contributors are celebrated researchers in these
domains.
New results and methods are presented in these contributions in order to foster
research in these areas in coming years. The reader will find survey presentations,
an account of recent development, and the exposition of new trends in such areas
from theoretical and applied perspectives.
As co-organizers of the workshop and editors of this volume it is our happy task
to thank those individuals and institutions whose efforts made it possible. First,
we acknowledge Universidad Sergio Arboleda for the financial and infrastructural
support. Second, it is a pleasure to thank all the members of the local organization
vii
viii PREFACE
Committee of the Universidad Sergio Arboleda for the excellent organization of this
meeting. Last, but certainly not least, we express our gratitude to the participants
of the seminar who made it a memorable event, to the contributors to this vol-
ume, and to Christine Thivierge of the AMS staff for her efficient support in the
publication of these proceedings.
Primitivo B. Acosta-Humánez, IMA - Universidad Sergio Arboleda

Francisco Marcellán, Universidad Carlos III de Madrid
Editors
Contemporary Mathematics
Volume 509, 2010
Differential Galois Theory of Algebraic Lie-Vessiot Systems
David Blázquez-Sanz and Juan José Morales-Ruiz
This paper is dedicated to Jairo Antonio Charris.
Abstract. In this paper we develop a differential Galois theory for algebraic

Lie-Vessiot systems in algebraic homogeneous spaces. Lie-Vessiot systems are
non autonomous vector fields that are linear combinations with time-dependent
coefficients of fundamental vector fields of an algebraic Lie group action. Those
systems are the building blocks for differential equations that admit superpo-
sition of solutions. Lie-Vessiot systems in algebraic homogeneous spaces in-
clude the case of linear differential equations. Therefore, the differential Galois
theory for Lie-Vessiot systems is an extension of the classical Picard-Vessiot
theory. In particular, algebraic Lie-Vessiot systems are solvable in terms of
Kolchin’s strongly normal extensions. Therefore, strongly normal extensions
are geometrically interpreted as the fields of functions on principal homoge-
neous spaces over the Galois group. Finally we consider the problem of inte-
grability and solvability of automorphic differential equations. Our main tool
is a classical method of reduction, somewhere cited as Lie reduction. We de-
velop and algebraic version of this method, that we call Lie-Kolchin reduction.
Obstructions to the application are related to Galois cohomology.
Contents
1. Introduction 2
2. Algebraic Groups and Homogeneous Spaces 4
3. Differential Algebraic Geometry 8
4. Galois theory of Algebraic Lie-Vessiot Systems 17
5. Algebraic Reduction and Integration 35
6. Integrability of Linear Equations 48
Appendix A. Stalk formula for affine morphisms 55
References 56
1991 Mathematics Subject Classification. Primary: 34M15 12H05; Secondary: 14L99, 34A26.
Key words and phrases. Differential Galois Theory, Differential Equations in the Complex
Domain.
This research has been partially supported by grant MCyT-FEDER MTM2006-00478 of
Spanish goverment, and the Sergio Arboleda University Research Agency CIVILIZAR.
1
2 DAVID BLÁZQUEZ-SANZ AND JUAN JOSÉ MORALES-RUIZ
1. Introduction
A Lie-Vessiot system, as defined in [3], is a system of non-autonomous differ-
ential equations,
(1.1) ẋi = Fi (t, x1 , . . . , xn ),
such that there exist r functions f (t) of the parameter t verifying:

r
Fi (t, x1 , . . . , xn ) = fj (t)(Aj xi ),
j=1
where A1 , . . . , As are autonomous vector fields which infinitesimally span a pre-

transitive Lie group action. Such systems were introduced by S. Lie at the end of
19th century (see, for instance [24]). The differential equation (1.1), interpreted
as a non-autonomous vector field, in a manifold M , is a linear combination of the
infinitesimal generators of the action of G in M :

X = ∂ + fj (t)Aj .
∂t
In [3], it is proven that a differential equation admits a superposition law if
and only if it is a Lie-Vessiot system related to a pretransitive Lie group action
(this is the global version of a classical result exposed in [24]). The orbits by a
pretransitive group action are homogeneous G-spaces, so that we can decompose
a Lie-Vessiot system in a family of systems on homogeneous spaces. Therefore,
Lie-Vessiot systems on homogeneous spaces are the building blocks of differential
equations admitting superpostion laws.
Here, we study Lie-Vessiot systems on algebraic homogeneous spaces M with
coefficients fi in a differential field K whose field of constants C is the field of
definition of the phase space M . In this frame, a Lie-Vessiot system is seen as a
derivation of the scheme MK , compatible with the canonical derivation of K.
We apologize with the reader not familiarized with the lenguaje of algebraic
geometry. We wish we will have a much simpler presentation of this results, but at
the actual state of research there is a number of technical points that seem to be
unavoidable.
Structure of the paper and original results. Section 2 is devoted to the ge-
ometry of algebraic groups and homogeneous spaces. Those results are well known
and can be found if references like [28] or [19] (for the Galois cohomoloy). In Sub-
section 2.6 we introduce an algebraic geometric interpretation of the fundamental
fields of the differential geometry.
In section 3 we expose the fundamentals of differential algebra and differential
algebraic geometry. That is, the systematic study of the set of prime differential
ideals of a differential algebra. We also use the languages of differential schemes
of Kovacic [21] and schemes with derivation of Buium [7]. This technical point is
necessary for dealing with differential equations in non-affine varieties. It makes the
theory a little more messy, but in the other hand these equations are ubiquitous:
let us consider for instance Riccati or Waierstrass equations. We also stablish a
dictionary between these languages, which is the Theorem 3.28.
Section 4 is devoted to the Galois theory of algebraic Lie-Vessiot systems. This
is the theoretical core of the paper. First of all, in Subsection 4.1 we discuss the
DIFFERENTIAL GALOIS THEORY OF ALGEBRAIC LIE-VESSIOT SYSTEMS 3
interpretation of differential equations as schemes with derivations, and their solu-

tions as points of differential schemes. We give the definition of rational solution
(Definition 4.2) and reduction (Definition 4.3). In Subsection 4.2 it is introduced
the definition of algebraic Lie-Vessiot system (Definition 4.5) in some technical
way; we encourage the reader to have a look on [3] and [4] for a more geometric
approach, and motivation. Those Lie-Vessiot systems are differential equations in
homogeneous spaces, and can be easily translated into differential equations in Lie
groups (automorphic systems, introduced by Vessiot [45]). We show that the inte-
gration of a Lie-Vessiot system is equivalent to that of its associated automorphic
system (Theorem 4.19). For the rest of the section we deal just with automorphic
systems. In order to do that, we use the logarithmic derivative, and automorphic
equations introduced in Subsections 4.3 and 4.4). Then we prove that, for a given
automorphic system there is a homogeneous space of Kolchin closed points. They
correspond to spliting fields (Galois extensions) of the automorphic system. We
show that they are isomorphic strongly normal extensions of the field of coeffi-
cients. We define the Galois group as the isotropy group of those points, and prove
that this group acts on the splitting field as its group of differential automorphisms.
What we have done is equivalent to the theory of G-primitive extensions, but using
a modern algebraic geometric point ow view. Finally, the Galois correspondence is
revisited under this geometric point of view, we prove that differential field exten-
sions spanned by solutions of Lie-Vessiot systems are Lie extensions in the sense of
Nishioka [31] (Theorem 4.45).
Section 5 deals with the problem of integrability, based in the Galois theory de-
veloped in the previous section. The practical tool we apply is a reduction method
of differential equations due to S. Lie, that we state in an differential algebraic
setting (Theorem 5.3). This method allow us to generalize to the automorphic
case the classical theorem of Kolchin and Kovacic of reduction of linear equations
(Theorem 5.8). Then we study the integrability by cuadratures of automorphic
equations, following some classical ideas that were formulated in algebraic terms by
Kolchin. Main idea is that, a quadature correspond to an elementary automorphic
equation in an abeliam group, depending on the nature of the group there appear
different kinds of quadratures. Subsections 5.3 to 5.6 survey them. Then we study
the notions of Liouville and Kolchin integrability, which are special kinds of inte-
grability by quadratures, and were deeply studied in [18]. Our approach allow us
to give an easy proof of Drach-Kolchin theorem on the ℘ elliptic function. The rest
of the section is manly due to Kolchin. The definition of quadrature imply that
the obstruction to integrablility is always contained in the non-abelian part of the
system. Since the non-lineal algebraic group are, by construction, abelian, then we
can use the structure of algebraic groups to linearize automorphic sistem. This is
seen in Subsections 5.9 to 5.12.
Section 6 is devoted to the problem of integrability by quadratures of linear
equations. We consider the hierarchy of Lie-Vessiot systems associated to an auto-
morphic system, and try to get information from the known solutions of equations
in the hierarchy. Main result is Theorem 6.2 which is a generalization of a classical
result of Liouville on the solutions of the Riccati equation. The rest of the paper
is devoted to the formulae and explicit forms of differential equations appearing in
the hierarchy and the specific example of the rigid movements that was treated by
Darboux in [10].
Finally there is an appendix devoted to the stalk formula for afine scheme
morphisms, as it is used in the proof of the strong normality of the Galois extensions.
Notation and Conventions. We denote differential and ordinary fields and

rings by calligraphic letters C, K, . . . The canonical derivation of a differential ring
K is denoted by ∂K or just ∂ whenever it does not lead to confussion. Algebraic
varieties are denoted by capital letters M, G, . . . The structure sheaf of M is denoted
by OM . If M is a C-algebraic variety and C ⊂ K, the space of K-points of an
algebraic variety M is denoted by M (K). We write MK for the K-algebraic variety
obtained after base change M ×C Spec(K). If p is a point of M we denote by κ(p)
its quotient field and p the valuation morphism p : OM,p → κ(p).
2. Algebraic Groups and Homogeneous Spaces

2.1. Algebraic Groups. Let us consider a field C and its algebraic closure C. ¯
By an algebraic variety over C we mean a reduced and separated scheme of finite
type over C. Along this text an algebraic group means an algebraic variety endowed
with an algebraic group law and inversion morphism. In particular, algebraic groups
over fields of characteristic zero are smooth varieties ([28] pp. 101–102).
The functor of points of an algebraic group takes values on the category of
groups. If G is a C-algebraic group, and K is a C-algebra, then the set G(K) of
K-points of G is naturally endowed with an structure of group.
An algebraic group is an affine group if it is an affine algebraic variety. The
main example of an affine algebraic group is the General Linear Group,
1
GL(n, C) = Spec (C[xij , ∆]) , ∆= .
|xij |
We call algebraic linear groups to the Zariski closed subgroups of GL(n, C). It
is well known that any affine algebraic group is isomorphic to an algebraic linear
group.
2.2. Lie Algebra of an Algebraic Group. Let us consider X(G) the space
of regular vector fields in G, id est, derivations of the sheaf OG vanishing on C. The
Lie bracket of regular vector fields is a regular vector field, so X(G) is a Lie algebra.
Definition 2.1. Let A be a regular vector field in G, and ψ : G → G an
automorphism of algebraic variety. Then, we define ψ(A) the transformed vector
field ψ(A) = (ψ )−1 ◦ A ◦ ψ .
ψ(A)
OG / OG
O
ψ (ψ )−1

OG
A / OG
Any C-point σ of G induces right and left translations, Rσ and Lσ , which are
automorphisms of the algebraic variety G. A C-point
¯ σ̄ of G, induces translations
in GC̄ .
Definition 2.2. The Lie algebra R(G) of G is the space of all regular vector
fields A ∈ X(G) such that for all C-point
¯ σ ∈ G(C),¯ Rσ (A ⊗ 1) = A ⊗ 1. In the
same way, we define the Lie algebra L(G) of left invariant vector fields.
The Lie bracket of two right invariant vector field is a right invariant vector
field. The same is true for left invariant vector fields, so R(G) and L(G) are Lie
sub-algebras of X(G). For a point x ∈ G its tangent space Tσ G is defined as the
space of C-derivations from the ring of germs of regular functions, OG,σ with values
in its quotient field κ(σ). It is a κ(σ)-vector space of the same dimentsion than G.
Any regualr vector field X in X(G), can be seen as a map σ → X σ ∈ Tσ (G). Let
us consider e the identity element of G. If C is algebraically closed, for any vector
v ∈ Te G there are unique invariant vector fields R ∈ R(G) and L ∈ L(G) such that

Re = Le = v (see [28] pp. 98–99).
2.3. Algebraic Homogeneous spaces.

Definition 2.3. Let G be a C-algebraic group. A G-space M is an algebraic
variety over C endowed with an algebraic action of G,
a
G ×C M −
→ M, (σ, x) → σ · x.
Let M be a G-space. Then for each extension C ⊂ K, the group G(K) acts on
the set M (K). Therefore it is a G(K)-set in the set theoretic sense. Given a point
x ∈ M its isotropy subgroup is an algebraic subgroup of G that we denote by Hx .
It is defined by equation Hx · x = x. Note that it is not necessary for x to be a
rational point.
The intersection of the isotropy subgroups of all closed points of M is a normal
algebraic subgroup HM G. The action of G is called faithful if HM is the identity
element {e}, and it is called free if for any rational point x, Hx = {e}. It is called
transitive if for each pair of rational points x, y ∈ M there is a σ ∈ G such that
σ · x = y; id est there is only one orbit.
Definition 2.4. Let us consider the induced morphism,
(a × Id) : G ×C M → M ×C M, (σ, x) → (σx, x)
then,
(1) M is an homogeneous G-space if (a × Id) is surjective.
(2) M is a principal homogeneous G-space if (a × Id) is an isomorphism.
If C is algebraically closed, an homogeneous G-space is simply a transitive G-
space and a principal homogeneous G-space is a free and transitive G-space. In
such case, any principal homogeneous G-space over is isomorphic to G.
2.4. Existence of quotients: Chevalley’s theorem. Let V ve a C-vector

space, and GL(V ) the group of linear transformations of V . It is an C-algebraic
group, and it acts algebraically on any tensor space over V . Given a tensor T
we call stabilizer subgroup of T to the group of linear transformations σ ∈ GL(V )
for whom there exist a scalar λ ∈ C such that σ(t) = λT . In other words, the
stabilizer subgroup of T is the isotropy subgroup of the line T spanned by T in
the projectivization of the tensor space.
Theorem 2.5 (Chevalley, see [14] p. 80). Let V be a C-vector space of finite
dimension, and let H ⊂ GL(V ) be an algebraic subgroup. There exist a tensor,
∗
T ∈ V ⊗ni ⊗C V ⊗mi
i
such that H is the stabilizer of T ,

H = {σ ∈ GL(V )|σ(T ) = T }
From this result we obtain that for a linear algebraic group G and an algebraic
subgroup H, the quotient space G/H is isomorphic to the orbit OT in the projec-
⊗n ∗
tive space P i V
i
⊗C (V ⊗mi ) . It is a quasiprojective algebraic variety.
There is a lack in the literature of an existence theorem for arbitrary quotients
of an non-linear algebraic group over an arbitrary field. However, there is a result,
due to M. Rosenlicht [33], saying that for any action of an algebraic group G on
an algebraic variety V , there exist a G-invariant open subset U ⊂ V such that the
geometrical quotient U/G in the sense of Mumford exists. In the case of a subgroup
G acting on G, this open subset must be right-invariant, and then it coincides with
G.
2.5. Galois Cohomology. In this section, we assume that C is a perfect field ;

note that this holds if C is of characteristic zero, which is the case we are interested
in. In such case, any algebraic extension can be embedded into a Galois extension.
Therefore, the algebraic closure C¯ is the inductive limit of all Galois extensions of C.
The group of C-automorphisms of C¯ is then identified with the projective limit of all
Galois groups, of algebraic extensions of C. With the initial topology of the family of
projections onto finite Galois groups, this is a compact totally disconnected group,
that we denote Gal(C/C).
¯
Let G be a C-algebraic group. The group of automorphisms acts on G(C) ¯ by
composition. Let us consider Gk the set of continuous maps from Gal(C/C) ¯ k onto
G(C).
¯ In such case G0 = G(C). ¯ We consider the sequence:
δ δ
(2.1) 0 → G0 −→
0
G1 −→
1
G2 ,
where the codifferential of x ∈ G0 is (δ0 x)(σ) = x−1 · σ(x), and the codifferential
of ϕ ∈ G1 is (δ1 ϕ)(σ, τ ) = ϕ(σ · τ )−1 · ϕ(σ) · σ(ϕ(τ )). An element in the image
of δ0 is called a coboundary, the set of coboundaries is denoted by B 1 (G, C). An
element ϕ ∈ G1 is called a 1-cocycle if δ1 ϕ vanish. The set of 1-cocycles is denoted
Z 1 (G, C). Two 1-cocycles are called cohomologous if there is x ∈ G0 such that
ϕ(σ) = x−1 · ψ(σ) · σ(x). This is an equivalence relation in Z 1 (G, C). The quotient
set Z1 (G, C)/ ∼ is a pointed set, with distingished point the class of coboundaries.
Note that when G is an abelian group the sequence (2.1) is a differential complex
and this quotient is the first cohomology group.
Definition 2.6. The zero Galois cohomology set of G with coefficients in C,
H 0 (G, C) is the kernel of δ0 . It is a pointed set with distinguised point the identity.
The first Galois cohomology set of G with coefficients in C, H 1 (G, C), is the pointed
set Z 1 (G, C)/ ∼.
From the definition of δ0 it is clear that x ∈ H 0 (G, C) if an only if it is invariant
under the action of Gal(C/C).
¯ The fixed field of C¯ in precisely C, therefore the
zero Galois cohomology set coincides with the set of C-points G(C). Therefore, we
define the zero cohomology set H 0 (V, C) of any C-algebraic variety V to be the set
of C-points V (C).
Let G be an algebraic subgroup of G. In such case H 0 (G/G , C) is a pointed
set, with distinguised point the class of the identity. An element x ∈ H 0 (G/G , C)
is a C-point of the homogeneous space G/G . This x is the class of a unique C- ¯

point x̄ of G. The coboundary ∂0 x̄ is a cocycle in G , and its cohomology class
[x̄] ∈ H 1 (G , C) does not depends on the election of x. We have a morphism of
pointed sets H 0 (G/G , C) → H 1 (G, C) called the connecting morphism. We obtain
an exact sequence of pointed sets:
0 → H 0 (G , C) → H 0 (G, C) → H 0 (G/G , C) → H 1 (G , C) → H 1 (G, C)
and when G is a normal subgroup of G, the sequence
H 1 (G , C) → H 1 (G, C) → H 1 (G/G , C)
is also exact (see [19], p. 277–288).
Using the previous exact sequence it is relatively easy to compute the first
Galois cohomology set of several algebraic groups. We say that the first cohomology
set of G with coefficients in C vanish if it consists of an only point. In particular
the following results are well known:
• The first cohomology set of the additive group H 1 ((C, +), C) vanish.
• The first cohomology set of the multiplicative group H 1 (C ∗ , ·), C) vanish.
• H 1 (GL(n, C), C) vanish.
• H 1 (SL(n, C), C) vanish.
• If G is linear connected solvable group then H 1 (G, C) vanish.
• If C is algebraically closed then for any algebraic group H 1 (G, C) vanish.
• If S is a Riemann surface and M(S) is its field of meromorphic function
then for any linear connected M(S)-algebraic group G, H 1 (G, M(S))
vanish (this is a particular case of fields of dimension lower or equal than
one, treated in [35]).
• If S is an open Riemann surface then for any connected M(S)-algebraic
group H 1 (G, M(S)) vanish (Grauert theorem, see [36]).
The first Galois cohomology set classifies the principal homogeneous spaces over
G. This classification was first obtained by Châtelet for some particular cases, here
we follow Kolchin [19] (see p. 281–283). The main fact is that if the first Galois
cohomology set vanish then all principal homogeneous spaces have rational points.
Theorem 2.7. Let G be a C-algebraic group and M a principal homogeneous G-
space. Then M defines a class [M ] in H 1 (G, C). This cohomology class classifies M
up to C-isomorphisms. M is isomorphic to G if and only if [M ] is the distinguised
point of H 1 (G, C). Reciprocally any cohomology class of H 1 (G, C) is the class of
certain homogeneous G-space.
2.6. Fundamental Fields. Consider a right invariant vector field A ∈ R(G).
Then, A ⊗ 1 is a regular vector field in G ×C M . This vector field is projectable by
the action of G in M ,
a : G ×C M → M, ⊗ 1 → A
A M .
Definition 2.8. We call algebra of fundamental field R(G, M ) to the Lie

M of vector fields
algebra of regular vector fields in M spanned by the projections A
A ⊗ 1, being A right invariant vector field in G.

There is a canonical surjective Lie algebra morphism,

R(G) → R(G, M ), →A
A M ,
the kernel of this morphism is the Lie algebra of the kernel of the action HM ,
R(HM ) ⊂ R(G). In particular, the Lie algebra of fundamental fields R(G, G) in G
coincides with R(G).
3. Differential Algebraic Geometry

We can state that the differential algebraic geometry is with respect to the
differential algebra the same than the classical algebraic geometry is with respect
to the commutative algebra. In this sense, the differential algebraic geometry is the
study of geometric objects associated with differential rings. Here we present the
theory of schemes with derivations, which has been developed by Buium [7], and
the theory of differential schemes, which is due to Keigher [16, 17], Carra’ Ferro
(see [9]), and Kovacic [21].
3.1. Differential Algebra. We present here some preliminaries in differential
algebra. The main references for this subject are [32], [15], [19].
A differential ring is a commutative ring A and a derivation ∂A . By a derivation
we mean an application verifying the Leibnitz rule, ∂A (ab) = a·∂A (b)+b·∂A (a). An
element a ∈ A is called a constant if it has vanishing derivative ∂a = 0. Whenever
it does not lead to confusion, we will write ∂ instead of ∂A . The subset CA of
constants elements is a subring of A. When A is a field we call it a differential
field. In such a case, the constant ring CA is a subfield of A. An ideal I ⊂ A is a
differential ideal if ∂(I) ⊂ I.
Note that if I is a differential ideal, then the quotient A/I is also a differential
ring. For a subset S ⊂ A we denote [S] for the smallest differential ideal con-
taining S, and {S} for the smallest radical differential ideal containing S. For an
ideal I ⊂ A we denote I for the smallest differential ideal containing I, namely:

I = i ∂ i (I). Localization by arbitrary multiplicative sytems is also suitable in
differential rings. A ring morphism is called differential if it is compatible with the
derivation. In the category of differential rings, tensor product is also well defined.
Consider K a differential field. A differential ring A endowed with a morphism
K → A is called a differential K-algebra. If A is a differential field then we say that
it is a differential extension of K.
√
3.2. Keigher Rings. If I ⊂ A is an ideal, we denote its radical ideal by I,
the intersection of all prime ideals containing I. In algebraic geometry, there is
a one-to-one correspondence between the set of radical ideals of A and the set of
Zariski closed subsets of Spec(A), the prime spectrum of A. In order to perform
an analogous systematical study of the set of differential ideals - id est differential
algebraic geometry - we should require radicals of differential ideals to be also
differential ideals. This property does not hold in the general case. We have to
introduce a suitable class of differential rings. This class was introduced by Keigher
(see [16]); we call them Keigher rings.
Definition 3.1. A Keigher √ ring is a differential ring verifying that for each
differential ideal I, its radical I is also a differential ideal.
Definition 3.2. For any ideal I ⊂ A we define its differential core as I =
{a ∈ I : ∀n(∂ n a ∈ I)}.
Keigher rings can be defined in several equivalent ways. The following theorem
of characterization includes different possible definitions (see [21], proposition 2.2.).
Theorem 3.3. Let A be a differential ring. The following are equivalent:

(a) If p ⊂ A is a prime ideal, then p is a prime differential ideal.
(b) If I ⊂ A is a differential ideal, and S is a multiplicative system disjoint
from I, then there is a prime maximal differential ideal containing I dis-
joint with S. √
(c) If I ⊂ A is a differential ideal, then so is I.
(d) If S is any subset, then {S} = [S].
(e) A is a Keigher ring.
By a Ritt algebra we mean a differential ring including the field Q of rational
numbers. When studing differential equations in characteristic zero, differential
rings considered are mainly Ritt algebras. A main property of Ritt algebras is
that the radical of a differential ideals is a differential ideal (see for instance [15]),
therefore Ritt algebras are Keigher rings.
Proposition 3.4. If A is a Keigher ring then for any differential ideal I, A/I
is Keigher and for any multiplicative system S, S −1 A is Keigher.
Proof. Assume A is Keigher. First, let us prove that A/I is Keigher. Consider
√
the projection π : A → A/I. Let a be a differential ideal of A/I. Then a =
π( π −1 (a)) is a differential ideal.
Second, consider a localization morphism l : A → S −1 A. Let a ⊂ S −1 A be a
differential ideal. Let us denote by b the preimage l−1 (a); it is a differential ideal
and l(b) · S −1 A = a.
√ √ n
Let us consider as ∈ a. as 1s = a1 ∈ a. For certain n, hence a1 ∈ a, an ∈ b and
√ √ m
a ∈ b. A is Keigher, and then ∂a ∈ b. Therefore (∂a)m ∈ b, so that ∂a ∈a
√ 1
and, for instance, 1 ∈ a. Finally,
∂a
a ∂a 1 a ∂s √
∂ = − ∈ a,
s 1 s 1 s2
and by (c) of Theorem 3.3 S −1 A is Keigher.
3.3. New Constants. From now on let K be a differential field, and let C be
its field of constants. We assume that C is algebraically closed. A classical lemma
of differential algebra (see [19] p. 87 Corollary 1) says that if A is a differential
K-algebra, then the ring of constant CA is linearly disjoint over C with K. Let us
set this classical lemma in a more geometric frame.
Lemma 3.5. Let A be an integral finitely generated differential K-algebra. Then
there is an affine subset U ⊂ Spec(A) such that the ring of constants CAU is a
finitely generated algebra over C.
Proof. Consider Q(A) the field of fractions of A. The extension K ⊂ Q(A) is
of finite transcendency degree. Then, K ⊂ K ·CQ(A) ⊂ Q(A) are extensions of finite
transcendency degree, and there are λ1 , . . . λs in CQ(A) such that K(λ1 , . . . , λs ) =
K · CQ(A) . Constants λ1 ,. . .,λs are fractions fgii . Consider the affine open subset
obtained by removing from Spec(A) the zeroes of the denominators,
s
U = SpecA \ (gi )0 .
i=1
Then, λi ∈ AU and K[CAU ] = K[λ1 , . . . , λs ]. We will prove that CAU = C[λ1 , . . . , λs ].

Let λ ∈ CAU . It is certain polynomial in the variables λi with coefficients in K:

λ= aI λI , aI ∈ K;
I∈Λ
where Λ is a suitable finite set of multi-indices. We can take this set in such way
that the {λI }I∈Λ are linearly independent over K, and then so they are over C.
{λ, λI }I∈Λ is a subset of K-linearly dependents elements of CAU . By [19] (p. 87
corollary 1) then they are C-linearly dependent. Hence, λ is C-linear combination
of {λI }I∈Λ , λ ∈ C[λ1 , . . . , λs ] and finally CAU = C[λ1 , . . . , λs ].
3.4. Differential Spectra.
Definition 3.6. Let A be a differential ring. DiffSpec(A) is the set of all prime
differential ideals p ⊂ A.
Let S ⊂ A any subset. We define the differential locus of zeroes of S, {S}0 ⊂
DiffSpec(A) as the subset of prime differential ideals containing S. This family
of subsets define a topology (having these subsets as closed subsets), that we call
the Kolchin topology or differential Zariski topology. Note that {S}0 = (S)0 ∩
DiffSpec(A). From that if follows:
Proposition 3.7. DiffSpec(A) with Kolchin topology is a topological subspace
of Spec(A) with Zariski topology.
From now on, let us consider the following notation: X = Spec(A), and X =
DiffSpec(A).
Let us recall that a topological space is said reducible if it is the non-trivial
union of two closed subsets. It is said irreducible if it is not reducible. A point
of an irreducible topological space is said generic if it is included in each open
subset. The following properties of the differential spectrum are proven in [17] (see
Proposition 2.1).
Proposition 3.8. X verifies:
(1) X is quasicompact.
(2) X is T0 separated.
(3) Every closed irreducible subspace of X admits a unique generic point.

The map X → 2X , that maps each point x to its Kolchin closure {x} is
a bijection between points of X and irreducible closed subspaces of X .
Here we review some of the topological properties of the differential spectrum
of Keigher rings.
Lemma 3.9. Assume that A is a Keigher ring. Then each minimal prime ideal
is a differential ideal.
Proof. Then, let p be a minimal prime. By Theorem 3.3 (a), p is a prime
differential ideal and p ⊆ p.
Proposition 3.10. Assume that A is Keigher. Then, X is an irreducible
topological space if and only if X is an irreducible topological space.
Proof. Just note that the irreducible components of X are the Kolchin clo-
sure of minimal prime ideals of A.
Proposition 3.11. Assume A is Keigher. If X is connected, then X is con-

nected.
Proof. Assume that X = Y Z, then we have an isomorphism of rings
(p1 , p2 ) : A → OX (Y ) × OX (Z), a → (a|X , a|Y ),
the kernel of each restriction pi is intersection of minimal prime ideals, so by Lemma
3.9 they are differential ideals. Hence, the rings OX (Y ) and OX (Z) are also differ-
ential rings. Then,
X = Y Z ,

being Y = DiffSpec(OX (Y )), Z = DiffSpec(OX (Z)). We have proven that if X
disconnects, then X disconnects.
3.5. Structure Sheaf. We define the structure sheaf OX as in [21]. Let us
consider the projection,
π: Ax → X .
x∈X
being x∈X Ax the disjoint union of all the localized rings Ax . We say that a
section s of π defined in an open subset U ⊂ X is a regular function if it verifies
the following: for all x ∈ U there exist an open neighborhood x ∈ Ux and a, b ∈ A
with b(x) = 0 (b ∈ x), such that for all y ∈ Ux with b(y) = 0, s(y) = ab ∈ Ay .
Thus, a regular function is a section which is locally representable as a quotient.
We write OX for the sheaf of regular functions in X . By the above construction
we can state:
Proposition 3.12. The stalk OX ,x is a ring isomorphic to Ax .
Theorem 3.13. Let us consider the natural inclusion j : X → X. The sheaf
of regular functions OX is the restriction OX |X of the sheaf of regular function in
X.
Proof. First, let us define a natural morphism of presheaves of rings on X
between the inverse image presheaf j −1 OX and OX . Let us consider an open
subset U ⊂ X and a section s of the presheaf j −1 OX defined in U . By definition
of inverse image, there is an open subset W of X such that W ∩ X ∩ U and for
what s is written as a fraction ab ∈ AW . This fraction is a section of OX (U ), and
it defines the presheaf morphism
j −1 OX → OX .
This presheaf morphism induces a morphism between associated sheaves OX |X
and OX . It is clear that this natural morphism induce the identity between fibers
(j −1 OX )x = Ax → OX ,x = Ax , and then it is an isomorphism.
3.6. Global Sections. One of the main facts of the differential algebraic ge-
ometry is that the ring of global regular sections of X does not coincide with the
differential ring A. Of course there is a canonical morphism from A to OX (X ).
However there are non-vanishing elements giving rise to the zero section and non
invertible elements giving rise to invertible sections. An element a of A is called
a differential zero if its annihilator ideal is not contained in any proper differen-
tial ideal. The set of differential zeroes is denoted by Z. An element is called a
differential unit if it is not contained in any proper differential ideal. The set of
differential units is denoted by U. Then, there is a canonical injective morphism,
U−1 A/Z → OX (X ). But in general this morphism is not surjective, id est, there
are regular functions that are not representable as fractions of A. Therefore, the
differential spectrum of OX (X ) is not always isomorphic to X . This problem is
extensively discussed in [2].
3.7. Differential Schemes. The study of differential schemes started within
the work of Keigher [16, 17] and was continued by Carra’ Ferro [9], Buium [7] and
Kovacic [21]. Definitions are slightly different in each author approach, here we
follow Kovacic.
Let us remind that a locally ringed space is a topological space X endowed with
an structure sheaf of rings OX such that for all x ∈ X the stalk OX,x is a local ring.
Thus, a locally differential ringed space is a locally ringed space whose structure
sheaf OX is a sheaf of differential rings. A morphism of locally differential ringed
spaces f : X → Y consist of a continous map together with a sheaves morphism
f : OX → f∗ OY . For the differential ring A it is clear that its differential spectrum
X endowed with the structure sheaf OX is a locally differential ringed space.
Definition 3.14. An affine differential scheme is a locally differentially ringed
space X which is isomorphic to DiffSpec(A) for some differential ring A.
Definition 3.15. A differential scheme is a locally differentially ringed space
X in which every point has a neighborhood that is an affine differential scheme.
Remark 3.16. Schemes are differential schemes, endowed with the trivial
derivation. The category of differential schemes is an extension of the category
of schemes, in the same way that the category of differential rings is an extension
of the category of rings.
By a morphism of differential schemes f : X → Y we mean a morphism of
locally ringed spaces, such that f : OY → f∗ OX is a morphism of sheaves of
differential rings.
Let K be a differential field. A K-differential scheme is a differential scheme
X provided with a morphism X → DiffSpec(K), it means that OX is a sheaf of
differential K-algebras.
A morphism of differential schemes f : X → Y between two differential K-
schemes is a morphism of differential K-schemes if the sheaf morphism f : OY →
f∗ OX is a morphism of sheaves of differential K-algebras.
3.8. Product of Differential Schemes. There is not a direct product in the
category of differential schemes relative to a given basic differential scheme. This
problem is discussed in [21]. However, in the case of differential schemes over a
differential field K we can construct the direct product by patching tensor products,
as it is usually done in algebraic geometry. Therefore,
DiffSpec(A) ×K DiffSpec(B) = DiffSpec(A ⊗K B).
Moreover, if X and Y are reduced differential K-schemes then X ×K Y is also
reduced (see [22] Proposition 25.2).
3.9. Split of Differential Schemes.
Definition 3.17. Let X be a differential scheme. Define the presheaf of rings
CX on X by the formula,
CX (U ) = COX (U) ,
for any open subset U ⊆ X.

From this definition it follows that CX is a sheaf of rings and its fiber CX,x is
isomorphic to the ring of constants COX,x . In particular, if X is a K-differential
scheme CX is a sheaf of CK -algebras.
Definition 3.18. We call space of constants of X, Const(X) to the locally
ringed space (X, CX ).
Definition 3.19. We say that X is an almost-constant differential scheme if
its space of constants Const(X) is a scheme.
Let X be an almost-constant scheme. Then, each open subset U ⊂ X is
also almost-constant. If Y is a reduced closed subscheme of X then Y is almost-
constant. In this way if Y is a locally closed reduced subscheme of X, then Y is
almost-constant.
Let K be a differential field, and C its field of constants.
Definition 3.20. A differential K-scheme X splits if there is a C-scheme Y
and an isomorphism of K-differential schemes,
∼
φ: X −
→ Y ×C DiffSpec(K).
The isomorphism φ is called an splitting isomorphism for X.
Proposition 3.21. If X is reduced and splits, then it is almost-constant and
∼
X−
→ Const(X) ×C DiffSpec(K).
Proof. [22] proposition 28.2.
3.10. Strongly Normal Extensions. Strongly normal extensions are intro-
duced by Kolchin [18]. They are differential field extensions whose group of auto-
morphisms admits an structure of algebraic group. This notion has been recently
characterized in terms of differential schemes by Kovacic [23]. This characteriza-
tion is more convenient for our presentation of differential Galois theory, so that we
will use it as a new definition.
Definition 3.22. K → L is a strongly normal extension if and only if the
differential scheme DiffSpec(L ⊗K L) splits. In such case denote Gal(L/K) to the
scheme Const(DiffSpec(L ⊗K L)).
Note that prime differential ideals of L ⊗K L whose quotient field is L, corre-
spond to K-automorphisms of L. If σ is a K-automorphism of L, the kernel of the
differential K-algebra morphism,
L ⊗K L → L, a ⊗ b → aσ(b),
is a prime differential ideal pσ . Then, the set of rational points of DiffSpec(L ⊗K L)
is naturally endowed with a group structure. This group structure descent to a
structure of C-algebraic group precisely when DiffSpec(L ⊗K L) splits. In such case
the space of constant Gal(L/K) is endowed with an structure of algebraic group.
This problem is axhaustively treated in [23].
This approach gives us a parallelism with Galois extensions in classical theory
of fields. Note that a field extension k → K is a Galois extension if and only if
Spec(K ⊗k K) = G ×k Spec(K) (see [34]). We also obtain the scheme structure of
the Galois group: it is the scheme of constants of DiffSpec(L ⊗K L).
3.11. Galois Correspondence for Strongly Normal Extensions. Let us

consider as above K ⊂ L a strongly normal extension of differential fields. To each
subgroup H ⊂ Gal(L/K) we assign the intermediate extension K ⊂ LH ⊂ L of
H-invariants. Reciprocally to each intermediate extension K ⊂ F ⊂ L we assign
the subgroup Gal(L/F) ⊂ Gal(L/K) of automorphisms of L that are differential
F-algebra automorphism. The Galois correspondence between closed subgroups
and intermediate extensions is first shown by Kolchin (see [18] and [19]).
Theorem 3.23. The maps
H → LH ⊂ L
from group subschemes of Gal(L/K) to intermediate differential extensions and
F → Gal(L/F) ⊂ Gal(L/K)
from intermediate differential extensions subgroup schemes, are bijective and in-
verse each other. The extension K ⊂ F is strongly normal if and only if Gal(L/F)
is a normal subgroup of Gal(L/K). In such case Gal(F/K) is isomorphic to the
quotient Gal(L/K)/Gal(L/F).
3.12. Lie Extensions. The algebraic differential approach to Lie-Vessiot sys-
tems, in terms of differential fields, was initiated by K. Nishioka [31]. He relates
the differential extensions generated by solutions of a Lie-Vessiot system with alge-
braic dependence on initial conditions; a concept introduced by H. Umemura [40]
in relation with the analysis of Painlevé differential equations. He also introduces
the notion of Lie extension, a differential field extension that carry the infinitesimal
structure of a Lie-Vessiot system. Here we review some of his results, in order to
relate them with the Galois theory of automorphic systems. Consider a differential
field K of characteristic zero with algebraically closed constant field C. Any consid-
ered differential extension of K is a subfield of certain fixed universal extension of
K.
Definition 3.24. We say that a differential extension K ⊂ R depends ratio-
nally on arbitrary constants if there exist a differential field extension K ⊂ M such
that R and M are free over K and R · M = M · CR·M .
For a differential extension K ⊂ L denote DerK (L) the space of derivations of
L that vanish over K. This space is a K-Lie algebra.
Definition 3.25. We say that a differential extension K ⊂ L is a Lie extension
if C = CL , there exists a C-Lie sub algebra g ⊂ DerK (L) such that [∂, g] ⊂ Kg, and
Lg = DerK (L).
Theorem 3.26 ([31]). Suppose that K is algebraically closed. Then every in-
termediate differential field of a strongly normal extension of K is a Lie extension.
3.13. Schemes with Derivation. In this section we present some facts of the
theory of schemes with derivations. This is mainly the point of view of [7]. However
we consider only regular derivations whereas A. Buium considers the more general
case of meromorphic derivations. Our purpose is to relate schemes with derivations
to differential schemes. Note that the regularity of the derivation is essential to
Theorem 3.28 below; hence it does not hold under Buium’s definition.
Let X be a scheme. A derivation ∂X of the structure sheaf OX is a law that
assigns to each open subset U ⊂ X a derivation ∂X (U ) of the ring OX (U ). This
law is assumed to be compatible with restriction morphisms.
Definition 3.27. A scheme with derivation is a pair (X, ∂X ) consisting of a

scheme X and a derivation ∂X of the structure sheaf OX .
Thus, a scheme with derivation is a scheme such that its structure sheaf is
a sheaf of differential rings. A morphism of schemes with derivation is a scheme
morphism such that induces a morphism of sheaves of differential rings.
Let K be a differential field. A K-scheme with derivation is a scheme with
derivation (X, ∂) together with a morphism (X, ∂) → (Spec(K), ∂). Thus, the
structure sheaf of X is a sheaf of differential K-algebras.
Let (X, ∂X ), (Y, ∂Y ) be two K-schemes with derivation. Then the direct product
X ×K Y admits the derivation ∂X ⊗ 1 + 1 ⊗ ∂Y . Then,
(X ×K Y, ∂X ⊗ 1 + 1 ⊗ ∂Y )
is the direct product of (X, ∂X ) and (Y, ∂Y ) in the category of schemes with deriva-
tion.
3.14. Differential Schemes and Schemes with Derivation.
Theorem 3.28. Given a scheme with derivation (X, ∂) there exist a unique
topological subspace X ⊂ X verifying
(1) X endowed with the structure sheaf OX |X and the derivation ∂|X is a
differential scheme. This differential scheme will be denoted Diff(X, ∂).
(2) For each open affine subset U ⊂ X, U ∩ X DiffSpec(OX (U ), ∂).
Furthermore, each morphism of schemes with derivation (X, ∂X ) → (Y, ∂Y ) induces
a morphism of differential schemes Diff(X, ∂X ) → Diff(Y, ∂Y ). The assignation
(X, ∂) Diff(X, ∂) is functorial.
Proof. If X is an affine scheme then the theorem holds, and
X = DiffSpec(OX (X)).
Let us consider the non-affine case. Let (X, ∂X ) be an scheme with derivation,
and let {Ui }i∈Λ be a covering of X by affine subsets. The ring of sections OX (Ui )
is a differential ring for al i ∈ Λ, and its spectrum Spec(O(Ui )) is canonically
isomorphic to Ui .
For each i ∈ Λ we take take Ui the differential spectrum DiffSpec(OX (Ui )),
which is a topological subspace of Ui . Then Ui ⊂ Ui ⊂ X. Let us define X =

i∈Λ Ui . Thus, X is a locally differential ringed space with the sheaf OX |X .
Let us prove that X is a differential scheme.
First, let us prove that Ui ∩X = Ui . By construction we have, Ui ⊂ Ui ∩X . Let
us consider x ∈ Ui ∩ X . It means that for certain j ∈ Λ, x ∈ Ui ∩ Uj , and x ∈ Uj ⊂
Uj . Let us consider an affine neighborhood Ux of x contained in such intersection.
Because the inclusion Ux → Uj , we have that x ∈ Ux = DiffSpec(OX (Ux )). Then
we have inclusions and restriction as follows:
Ux A / Ui OX (Ux ) o OX (Ui ) Ux / U
AA eKKK @@ i
AA KKK @@
AA KKK @@
K @
Uj OX (Uj ) Uj
We conclude that x ∈ Ui .
Secondly, let us prove that for any affine subset U , the intersection U ∩ X is an
affine differential scheme DiffSpec(OX (U )). Let U be an affine subset, and let us
denote U the differential spectrum DiffSpec(OX (U )) that we consider as a subset

of U . Let us consider x ∈ U . Then, for certain i ∈ Λ, x ∈ U ∩ Ui . Let Ux be
an affine neighborhood of x such that Ux ⊂ U ∩ Ui . Denote by Ux the differential
spectrum of OX (Ux ). We have that Ux ⊂ Ui , and then x ∈ U ∩ X . Reciprocally
let us consider x ∈ U ∩ X . Then for certain i ∈ Λ we have x ∈ Ui . By the same
argument, we have that x ∈ U is a prime differential ideal of OX (U ).
The derivation ∂ induces derivations on the structure sheaf of U ∩ X for each
affine open subset U ⊂ X. Then, it induce a derivation ∂ : OX → OX and
Diff(X, ∂) = (X , OX |X , ∂|X ) is a differential scheme.
Finally, let us consider f : (X, ∂X ) → (Y, ∂Y ) a morphism of schemes with
derivation. If we assume that they are both affine schemes, then the theorem holds.
In the general case, we cover Y by affine subsets {Ui }i∈Λ , and each fiber f −1 (Ui )
by affine subsets {Vij }i∈Λ,j∈Π . Then f is induced by the family of differential ring
morphisms

fij : OY (Uj ) → OX (Vij ).
These morphisms induce morphisms,

fij : Vij → Ui ,
of locally differential ringed spaces which coincide on the intersections, and then
they induce a unique morphism,
f : X → Y .

Definition 3.29. Let (X, ∂) be an scheme with derivation. We will say that
x ∈ X is a differential point if x ∈ Diff(X, ∂).
Corollary 3.30. Let us consider (X, ∂) an scheme with derivation, and x a
point of X. Then; the following are equivalent:
(a) x ∈ X is a differential point.
(b) For each affine neighborhood U , x correspond to a differential ideal of
OX (U ).
(c) The maximal ideal mx of the local ring OX,x is a differential ideal.
(d) The derivation ∂ induces a structure of differential field in quotient field
κ(x).
(e) The derivation ∂ restricts to the Zariski closure of x.
3.15. Split of Schemes with Derivation. Let Z be a scheme provided with
the zero derivation. Then we will write Z instead of the pair (Z, 0). Consider a
differential field K and let C be its field of constants.
Definition 3.31. We say that a K-scheme with derivation (X, ∂) splits, if there
is a C-scheme Y , and an isomorphism
∼
φ : (X, ∂) −
→ Y ×C (Spec(K), ∂),
φ is called a splitting isomorphism for (X, ∂).
Definition 3.32. The space of constants Const(X, ∂) is locally ringed space
defined as follows: it is the topological subspace of differential points of X, endowed
with restriction of the sheaf of constant regular functions.
Proposition 3.33. Suppose (X, ∂) is Keigher, then

Const(X, ∂) = Const(Diff(X, ∂)).
Proof. As topological subspaces of X they coincide by construction. Let
X = Diff(X, ∂). If X is Keigher then OX (U ) = lim → OX (V ) (see [9]). And
U ⊆V
because of that we have,

C lim OX (V )
→
= lim
→
COX (V ) ,
U⊆V U⊆V
and we finish.
Definition 3.34. (X, ∂) is almost-constant if Const(X, ∂) is a scheme.

Proposition 3.35. If (X, ∂) splits, then Diff(X, ∂) splits. If (X, ∂) is reduced
and split, then it is almost-constant and
∼
(X, ∂) −
→ Const(X, ∂) ×C (Spec(K), ∂).
Proof. Let us consider the splitting isomorphism (X, ∂) → Y ×C (Spec(K), ∂).
It is clear that Diff(Y ×C (Spec(K), ∂)) = Y ×C DiffSpec(K). Then the above
splitting isomorphism induces the splitting isomorphism of the differential scheme
Diff(X, ∂). If X is reduced, then Diff(X, ∂) is also reduced, and then we apply
Proposition 3.21.
4. Galois theory of Algebraic Lie-Vessiot Systems

In this chapter we discuss the Galois theory of Lie-Vessiot systems on algebraic
homogeneous spaces. The field of functions of the independent variable is here
a differential field K of characteristic zero and with a field of constants C that
we assume to be algebraically closed. We modelize algebraic Lie-Vessiot systems
with coefficients in K as certain K-schemes with derivation. We study the general
solution of algebraic Lie-Vessiot systems. It means that we study the differential
extensions of K that allow us to split the Lie-Vessiot system, and the associated
automorphic system. We find that they are strongly normal extensions in the
sense of Kolchin [18], and then we can apply Kovacic’s approach to Kolchin’s
differential Galois theory. In fact, the Galois theory presented here should be seen
as a generalization of the classical Picard-Vessiot theory, obtained by replacing the
general linear group by an arbitrary algebraic group. However, the particular case
of Picard-Vessiot theory contains all obstructions to solvability, because the non-
linear part of an algebraic group over C is an abelian variety: abelian groups do not
give obstruction to integration by quadratures.
4.1. Differential Algebraic Dynamical Systems. Here we establish a par-

allelism between dynamical systems and differential algebraic terminology. From
now on let us consider a differential field K, and C its field of constants. We
assume that C is algebraically closed and of characteristic zero. We modelize non-
autonomous dynamical systemas as schemes with derivation. The phase space is
an algebraic variety M over the constant field C, and the extended phase space is
MK = M ×C Spec(K). Therefore, non-autonomous dynamical system on M with
coefficients in K is a derivation on MK .
Definition 4.1. A differential algebraic dynamical system is a K-scheme with

derivation (M, ∂M ) such that M is an algebraic variety over K. We say that (M, ∂M )
is non-autonomous if K is a non-constant differential field.
There is a huge class of dynamical systems that can be seen as differential
algebraic dynamical systems, as polynomial or meromorphic vector fields. It in-
cludes Lie-Vessiot systems in algebraic homogeneous spaces, hence it also includes
systems of linear differential equations. Furthermore, a differential algebraic study
of a dynamical system is suitable in the most general case, but results depend on
the choice of an adequate differential field K.
For a differential algebraic dynamical system (M, ∂M ) we have the associated
differential scheme Diff(M, ∂M ). As a topological space this differential scheme
is the set of all irreducible algebraic invariant subsets of the dynamical system.
By algebraic, we mean that they are objects defined by algebraic equations with
coefficients in K.
Let us recall that for a K-algebra L we denote by M (L) the set of L-points of
M . This sets consist of all the morphisms of K-schemes from Spec(L) to M , or
equivalently, of all the rational points of the extended scheme
ML = M ×K SpecL.
Definition 4.2. Let (M, ∂M ) be a K-scheme with derivation. We call rational
solution of (M, ∂M ) any rational differential point x ∈ Diff(M, ∂M ). Let us consider
a differential extension K ⊂ L. A solution with coefficients in L is an L-point
x ∈ M (L) such that the morphism
x : (Spec(L), ∂) → (M, ∂M ),
is a morphism of schemes with derivation. In such a case the image x(0) = x of the
ideal (0) ⊂ L by x is a differential point x ∈ Diff(M, ∂M ) and its quotient field κ(x)
is an intermediate extension,
K ⊂ κ(x) ⊂ L,
we say that κ(x) is the differential field generated by x ∈ M (L).
As in classical algebraic geometry, there is a one-to-one correspondence between
solutions with coefficients in L of (M, ∂M ) and rational solutions of the differential
algebraic dynamical system after a base change, (M, ∂M ) ×K (Spec(L), ∂).
Definition 4.3. Let us consider two differential algebraic dynamical systems
over K, (M, ∂) and (N, ∂). We say that (M, ∂) reduces to (N, ∂) if there is an
algebraic variety Z over C and,
(M, ∂) = (N, ∂) ×C Z.
The notion of reduction is a generalization of the notion of split. In particular,
to split means reduction to (Spec(K), ∂).
Given a differential algebraic dynamical system; what does it mean to integrate
the dynamical system? As algebraists, we shall use this term for writing down
the general solution of the dynamical system by terms of known operations, mainly
algebraic operations and quadratures. However, in the general context of dynamical
systems there is not a general definition for integrability. We are tempted to say that
integrability is equivalent to split. Notwithstanding, there are several situations in
which the general solution can be given, but there is not a situation of split. For
example, algebraically completely integrable Hamiltonian systems [1]. In such cases
the flux is tangent to a global lagrangian bundle, and the generic fibers of this bundle
are affine subsets of abelian varieties. It allows us to write down the global solution
by terms of Riemann theta functions and Jacobi’s inversion problem. However,
this general solution can not be expressed in terms of the splitting of a scheme with
derivation.
Split is the differential algebraic equivalent to Lie’s canonical form of a vector
field. The scheme with derivation Z ×C (Spec(K), ∂) should be seen as an extended
phase space, and ∂ as the derivative with respect to the time parameter. The
splitting morphism,
(M, ∂) → Z ×C (Spec(K), ∂),
can be seen as Lie’s canonical form, usually referred to, in dynamical system argot,
as the flux box reduction. Then Z is simultaneously the algebraic variety of initial
conditions, and the space of global solutions of the dynamical system. Our conclu-
sion is that the split differential algebraic dynamical systems are characterized by
following the property: its space of solutions is parameterized by a scheme over the
constants.
In the context of algebraic Lie-Vessiot systems we will see that algebraic solv-
ability of the problem, is equivalent to the notion of split (Theorem 4.19). And
then, this notion plays a fundamental role in our theory. We will see that generi-
cally, a Lie-Vessiot equation does not split. If we want to solve it, then we need to
admit some new functions by means of a differential extension of K ⊂ L. Thus, the
dynamical system splits after a base change to L. The Galois theory will provide
us with the techniques for obtaining such extensions and studying their algebraic
properties (Proposition 4.24).
4.2. Algebraic Lie-Vessiot Systems. From now on we will consider a fixed
characteristic zero differential field K whose field of constants C is algebraically
closed. Let G be a C-algebraic group, and M a faithful homogeneous G-space.
in M with coef-
Definition 4.4. A non-autonomous algebraic vector field X
ficients in K is an element of the vector space X(M ) ⊗C K.
A non-autonomous algebraic vector field X in M is written in the form,
s
=
X i,
fi X
i=1
for certain elements fi ∈ K and X

i ∈ X(M ). We define the derivation ∂ associated
X

to X as the following derivation of the extended scheme MK :
s
∂X : K ⊗C OM → K ⊗C OM , a ⊗ f → ∂a ⊗ f + (afi ⊗ X
i f ).
i=1
Definition 4.5. A non-autonomous algebraic vector field X in M with coef-

ficients in K is called a Lie-Vessiot vector field if belongs to R(G, M ) ⊗C K. The
differential algebraic dynamical system (MK , ∂X ) is called a Lie-Vessiot system in
M with coefficients in K.
The group G is, in particular, a faithful homogeneous G-space. Let us recall
that the Lie algebra of fundamental fields on the group G coincides with the Lie
algebra of right invariant vector field R(G). Then, a Lie-Vessiot vector field in G
with coefficients in K is an element of R(G) ⊗C K.
Definition 4.6. We call automorphic vector fields to the Lie-Vessiot vector

in G with coefficients in K is an element
fields in G. An automorphic vector field A
of R(G) ⊗C K.
The canonical isomorphism between R(G) and R(G, M ) allows us to translate
Lie-Vessiot vector fields in M to automorphic vector fields in G.
Definition 4.7. We call automorphic system associated to (M, ∂X ) to the
is the automorphic vector field whose corre-
Lie-Vessiot system (GK , ∂A ), where A
sponding Lie-Vessiot vector field in M is X.
From now on let X be a Lie-Vessiot vector field in M , with coefficients in K,

and let A be the associated automorphic vector field in G.
4.3. Logarithmic Derivative. A K-point of the algebraic group G has co-

efficients in a differential field, so that it can be differentiated. The derivative of a
K-point of G gives a tangent vector at a K-point of GK . If we translate this tan-
gent vector to a right invariant vector field, we obtain the logarithmic derivative.
In order to do so we identify systematically the Lie algebra R(G) with the tangent
space Te G = DerC (OG,e , C). It is also important to remark that the tangent space
is compatible with extensions of the base field in the following way:
∼
R(G) ⊗C K −
→ Te (GK ) = DerK (OGK ,e , K).
In classical algebraic geometry it is assumed that derivations of Te (GK ) vanish on
K. However, automorphic systems are by definition compatible with the derivation
∂ of K. Thus, the restriction of an automorphic vector field ∂A to e ∈ GK is not
a tangent vector of Te (GK ): it is shifted by ∂. We have identifications of K-vector
spaces:
∼ −∂
R(G) ⊗C K / R(G) ⊗C K + ∂ / Te (GK )

A / ∂ = ∂ + A
/A
e
A
Let us consider σ ∈ G(K) and the canonical morphism σ of taking values in

σ:
σ : OGK ,σ → K, f → f (σ).
Let us remember that there is a canonical form of extension of the derivation
∂ in K to a derivation in GK . We consider the direct product G ×C (Spec(K), ∂)
in the category of schemes with derivation. By abuse of notation we denote by ∂
this canonical derivation in GK . By construction we have that (GK , ∂) splits – the
identity is the splitting morphism – and Const(GK , ∂) = G. Let us consider the
following non-commutative diagram,
(4.1) OGK ,σ σ /K .
∂ ∂

OGK ,σ σ /K
Lemma 4.8. The commutator σ = [∂, σ ] of the diagram (4.1) is a derivation

vanishing on K, and then σ belong to the tangent space Tσ (GK ) (id est, the space
of derivations DerK (OGK ,σ , K)).
Proof. [∂, σ ] is the difference between two derivations, and then it is a deriva-
tion. Let us consider f ∈ K ⊂ OGK σ , then σ (f ) = ∂f − ∂f = 0.
If σ is a geometric point of GK , then Rσ−1 is a automorphism of GK sending σ

to e. It induces an isomorphism between the ring of germs OGK ,σ and OGK ,e , and
then an isomorphisms between the corresponding spaces of derivations:

Rσ
Tσ (GK )
−1
/ Te (GK ) R(G) ⊗C K
Definition 4.9. Let σ be a geometric point of GK ; we call logarithmic de-

rivative of σ, l∂(σ), to the automorphic vector fiel Rσ −1 ([∂, σ ]). The logarithmic
derivative is then a map:
l∂ : G(K) → R(G) ⊗C K.
Proposition 4.10. Properties of logarithmic derivative:
(1) Logarithmic derivative is functorial in K; for each differential extension
K ⊂ L we have a commutative diagram:
G(K) / R(G) ⊗C K

G(L) / R(G) ⊗C L
(2) Let us consider σ and τ in G(K):

l∂(στ ) = l∂(σ) + Adjσ (l∂(τ ))
(3) Let us consider σ ∈ G(K):
l∂(σ −1 ) = −Adjσ (l∂(σ)).
Proof. (1) comes directly from the differential field extension, (2) comes from
the right invariance, and (3) is corollary to (2).
4.4. Automorphic Equation.

Theorem 4.11. Let us consider K ⊂ L a differential extension. Then σ ∈ G(L)
is a solution of the differential algebraic dynamical system (GK , ∂A ) if and only if

l∂(σ) = A.
Proof. Let us consider σ ∈ G(L), and let B be its logarithmic derivative. The
space R(G) ⊗C L is canonically identified with the Lie algebra of right invariant
vector fields on the base extended L-algebraic group GL :
R(G) ⊗C L = R(GL ).
is seen as a derivation B
By this identification, the automorphic vector field B
of the structure sheaf OGL . The germ B(σ) at σ of B is a derivation of the ring

OGL ,σ . The composition with σ give us the tangent vector B

σ ∈ Tσ (GL ):

B
OGK ,σ S
(σ)
/O
SSSS GK ,σEE
SSSS E
SSSS EEEσE
SSSSEE
S")
σ
B
K
The value of B at the identity point is, by definition, l∂(σ). Since B is a right

invariant vector field we have l∂(σ) = Rσ−1 (Bσ ) = σ ◦ B(σ) ◦ Rσ−1 hence B
σ is
equal to the commutator [∂, σ ] of Definition 4.9. Then, B (σ) is the defect of the
diagram (4.1); therefore the following diagram commutes:
OGK ,σ σ /K .
(σ)
∂+B ∂

OGK ,σ σ /K
Furthermore, B is determined by the commutator B σ = [∂, σ ] and then it is unique

right invariant vector field in GL that forces the diagram to commute.
Let us note that the commutation of the above diagram holds if and only if the
kernel mσ of σ is a differential ideal. Then B is the unique right invariant vector
field in GL such that the maximal ideal mσ is a differential ideal. Let us note also
that, this derivation ∂ + B σ is the germ in σ of the automorphic derivation

∂B = ∂ + B,
the logarithmic derivative of σ, is the unique element of
we conclude that B,
R(G) ⊗C L such that σ is a differential point of (GL , ∂B ).
for the so-called

Because of that we can substitute the automorphic system A,
automorphic equation:
(4.2)
l∂(x) = A
4.5. Solving Lie-Vessiot Systems.

Definition 4.12. Let us consider σ ∈ G(K). We call gauge transformation
induced by σ to the left translation Lσ : GK → GK .
Lemma 4.13. (GK , ∂A ) splits if and only if the automorphic equation (4.2) has
at least one solution in G(K).
Proof. Assume (GK , ∂A ) splits. Let us consider the splitting isomorphism
ψ : (GK , ∂A ) → Z ×C (Spec(K), ∂).
Let x be a C-rational point of Z. Let us denote by xK the corresponding K-point
of GK obtained after the extension of the base field. Thus, ψ −1 (xK ) is a solution of
(4.2). Reciprocally, let us assume that there exists a solution σ of (4.2) in G(K).
Let us consider the gauge transformation:
Lσ−1 : GK → GK .
It applies σ onto the identity element e ∈ GK . But the logarithmic derivative l∂(e)
vanishes, so that Lσ−1 transforms ∂A into the canonical derivation ∂. We conclude
that Lσ−1 is an splitting isomorphism.
Lemma 4.14. Assume that (GK , ∂A ) splits. In such case we can choose the
splitting isomorphism between the gauge transformations of GK . This gauge trans-
formation induces the split of any associated Lie-Vessiot system (MK , ∂X ).
Proof. We use the same argument as above. If it splits,
s : (GK , ∂A ) → G ×C (Spec(K), ∂) = (G, ∂),
then the preimage of the identity element s−1 (e) = σ is a solution of the auto-
morphic system. So that the gauge transformation Lσ−1 : σ → e maps solutions
of (GK , ∂A ) to solutions of (GK , ∂) and it is an splitting isomorphism. For any
associated Lie-Vessiot system (MK , ∂X ), and any point x0 ∈ M (C) we have that
Lσ (x0 ) is a solution of (MK , ∂X ). So that Lσ sends solutions of the canonical
derivation ∂ to solutions of ∂X . Thus, its inverse Lσ−1 is an splitting isomorphism
for (MK , ∂X ).
Lemma 4.15. Let Z be a C-algebraic variety and (ZK , D) a non-autonomous

differential algebraic dynamical system over K. If (ZK , D)
splits then (ZK , D)
is
almost-constant and Const(ZK , D) Z.

splits. It implies that there exist an C-scheme

Proof. Assume that (ZK , D)
Y , such that ZK = Y ×C Spec(K). We have that ZK YK , and then Z Y .
Lemma 4.16. Let Z be a reduced C-scheme. There is a one-to-one correspon-

dence between closed subschemes of Z and closed subschemes with derivation of
(ZK , ∂) = Z ×C (Spec(K), ∂).
Proof. First, let us consider the affine case. Assume Z = SpecR for a C-
algebra R. The ring of constants CR⊗C K is R itself. It follows that Const(ZK , ∂) =
Z. It is clear that R ⊗C K is an almost-constant ring: each radical differential ideal
is generated by constants. Because of that there is an one-to-one correspondence
between radical ideals of R and radical differential ideals of K.
In the non-affine case, let us consider Y a closed sub-C-scheme of Z. The
canonical immersion (YK , ∂) ⊂ (ZK , ∂) identifies Y with a closed sub-K-scheme
with derivation of (ZK , ∂). Reciprocally, let (Ỹ , ∂|Ỹ ) be a closed sub-K-scheme
with derivation of (ZK , ∂). Let us consider {Ui }i∈Λ an affine covering of Z. The
collection {Vi }i∈Λ with Vi = Ui ×C K is then an affine covering of ZK . Each
intersection Ỹi = Ỹ |Vi is an affine closed sub-K-scheme of Vi . We are in the affine
case: by the above argument there are closed sub-C-schemes Yi ⊂ Ui such that
(Ỹi , ∂|Ỹi ) = Yi ×C (Spec(K), ∂). This family defines a covering of a closed sub-C-
scheme Y = i∈Λ Yi of Z.
Lemma 4.17. Let Z be a C-algebraic variety and (ZK , D) a non autonomous

algebraic dynamical system over K. Let Y ⊂ Z a locally closed subvariety, and
assume that D is tangent to Y , so that (YK , D|
Y ) is a sub-K-scheme with derivation.
splits then (YK , D|
If (ZK , D) Y ) splits.
Proof. By substituting Z for certain open subset we can assume that Y is

closed. Let us consider the splitting isomorphism,
→ Z ×C (Spec(K), ∂).
ψ : (ZK , D)
Y ) is a locally closed subscheme with derivation of Z×C (Spec(K), ∂).
The image ψ(YK , D|
By Lemma 4.16 it splits.
Lemma 4.18. Assume that the action of G on M is faithful. Then (GK , ∂A )
splits if and only if (MK , ∂X ) splits.
Proof. Lemma 4.14 says that if (GK , ∂A ) splits, then (MK , ∂X ) splits. Recip-
rocally, let us assume that (MK , ∂X ) splits. For each positive number r we consider
r r
the natural lifting to the cartesian power (MK , ∂X
). The splitting of (MK , ∂X
) in-
duces the splitting of those cartesian powers differential algebraic dynamical system
). For r big enough there is a point x ∈ M such that its orbit Ox is a
r r r
(MK , ∂X
principal homogeneous space isomorphic to G. Then (Ox,K , ∂X ) is a locally closed
r r
sub-K-scheme with derivation of (MK , ∂X
). By Lemma 4.17 it splits. We also know
that (Ox,K , ∂X ) is isomorphic to (GK , ∂A ). Finally, (GK , ∂A ) splits.
Theorem 4.19. Assume that the action of G on M is faithful. Then the

following are equivalent.
(1) The automorphic equation (4.2) has a solution in G(K)
(2) (GK , ∂A ) splits.
(3) There is a gauge transformation of GK sending A to 0.
(4) (MK , ∂X ) splits.
(5) (GK , ∂A ) splits, is almost-constant, and Const(GK , ∂A ) G.
(6) (MK , ∂A ) splits, is almost-constant, and Const(MK , ∂X ) M .
Proof. Equivalence between (1) and (2) comes from Lemma 4.13. Equivalence
between (2) and (3) comes from Lemma 4.14. (2) and (4) are equivalent by Lemma
4.18. By Lemma 4.15, they all imply (5) and (6).
4.6. Splitting Field of an Automorphic System. Note that a differential

extension K ⊂ L, induces a canonical inclusion,
R(G, M ) ⊗C K ⊂ R(G, M ) ⊗C L;
so that a Lie-Vessiot vector field with coefficients in K is a particular case of a
Lie-Vessiot vector field with coefficients in L. So that if (MK , ∂X ) is a Lie-Vessiot
system, then (ML , ∂X ) makes sense.
Definition 4.20. We say that a differential extension K ⊂ L is a splitting
extension for (MK , ∂X ) if (ML , ∂X ) splits.
From theorem 4.19, we know that K ⊂ L is a splitting extension of (MK , ∂X ) if
and only it is a splitting extension of (GK , ∂A ). Then we will center our attention
in the automorphic vector field A.
4.7. Action of G(C) on GK . For each σ ∈ G(C), Rσ is an automorphism of

GK . The composition law is an action of G on GK by the right side,
GK ×C G → GK .
The vector field A is right invariant, so that we expect the differential points of
(GK , ∂A ) to be invariant under right translations. In fact, the above morphism is a
morphism of schemes with derivation,
(GK , ∂A ) ×C G → (GK , ∂A ).
We apply the functor Diff, and then we obtain an action of the C-algebraic group
G on the differential scheme Diff(GK , ∂A ),
Diff(GK , ∂A ) ×C G → Diff(GK , ∂A ).
Assume that (GK , ∂A ) split. In such case, when we apply the functor Const to the
previous morphism, we obtain a morphism of schemes,
Const(GK , ∂A ) ×C G → Const(GK , ∂A ).
Because of the split we already knew that Const(GK , ∂A ) is a C-scheme isomorphic
to G. Furthermore, the above morphism says that the action of G by the right side
on this G-scheme is canonical. We have proven the following:
Lemma 4.21. Assume that (GK , ∂A ) splits. Then Const(GK , ∂A ) is a principal
G-homogeneous space by the right side.
4.8. Existence and Uniqueness of the Splitting Field.
Lemma 4.22. There is a differential point x ∈ Diff(GK , ∂A ) which is closed in
the Kolchin topology.
Proof. Let us consider the generic point p0 ∈ GK . In particular it is a differ-
ential point p0 ∈ Diff(GK , ∂A ). If p0 is Kolchin closed, then we finish and the result
holds. If not, then the Kolchin closure of p0 contains a differential point point p1
such that p0 specializes on it p0 → p1 . We continue this process with p1 . As GK
is an algebraic variety, and then a noetherian scheme, this process finish in a finite
number of steps and lead us to a Kolchin closed point.
Lemma 4.23. Let x ∈ Diff(GK , ∂A ) be a closed differential point. Then its field
of quotients κ(x) is a differential extension of K with the same field of constants;
Cκ(x) = C.
Proof. Reasoning by reductio ad absurdum let us assume that there exists
c ∈ Cκ(x) not in C. Let us consider an affine open neighborhood U of x and denote
by A its ring of regular functions. We identify x with a maximal differential ideal
x ⊂ A. Denote by B the quotient ring A/x. B is a differential subring of the
differential field κ(x). By Lemma 3.5 there exist b ∈ B such that the ring constants
CBb – of the localized ring Bb – is a finitely generated C-algebra. By reducing
our original neighborhood U – removing the zeros of b – we can assume that b
is invertible and then the localized ring Bb is just B. CB is a non-trivial finitely
generated C-algebra over C, because it contains an element c not in C. So that
there is a non-invertible element c2 ∈ CB . The principal ideal (c2 ) is a non trivial
differential ideal in B. Let us consider a regular function a2 such that a2 (x) = c2 .
Then ∂A a2 ∈ x and (a, x) is a non-trivial differential ideal of A strictly containing
x. We arrive to contradiction with the maximality of x.
Proposition 4.24. Let x ∈ Diff(GK , ∂A ) be a closed point. Then K ⊂ κ(x) is

a splitting extension of (GK , A).
Proof. Let x be a closed point. Then the canonical morphism x of taking

values in x, x : OGK ,x → κ(x) is a morphism of differential rings. Let U be an affine
neighborhood of the image of π(x) by the canonical projection π : GK → G. By
composition we construct a morphism Spec(κ(x)) → U ,
OG (U )
σ / κ(x) .
nnn7
x nnn

π
nnnnn
nn
OGK ,x
The morphism σ is the dual of a morphisms σ from Spec(κ(σ)) to U . In other

words, σ is a point of G(κ(x)). We consider σ as a rational differential point of
(Gκ(x) , ∂A ), and then it is a solution of the automorphic equation. By Lemma 4.13,
(Gκ(x) , ∂A ) splits.
Definition 4.25. We say that σ, as defined in the above proof, is the funda-
associated with the closed differential point x.
mental solution of A
Let us consider the action of G on GK by right translations. The derivation
∂A is invariant by right translations, and then it is a morphism of schemes with
derivation:
(GK , ∂A ) ×C G → (GK , ∂A )
We apply the functor Diff, thus we obtain a morphism of differential schemes which
is an algebraic action of G on the set of differential points.
Diff(GK , ∂A ) ×C G → Diff(GK , ∂A )
Proposition 4.26. The action of G(C) on the set of closed points of Diff(GK , ∂A )
is transitive.
Proof. Let us consider a Kolchin closed point x ∈ Diff(GK , ∂A ). Let L be the
rational field of x. It is an splitting field for (GK , ∂A ). We have that (GL , ∂A ) splits,
hence Diff(GL , ∂A ) is an almost-constant differential scheme. Thus Diff(GL , ∂A ) is
homeomorphic to the principal homogeneous G-space Const(GL , ∂A ). The differ-
ential extension K ⊂ L induces a commutative diagram of schemes with derivation,
(GL , ∂A ) ×C G / (GL , ∂ )
A
π1

(GK , ∂A ) ×C G / (GK , ∂ )
A
and thus, a commutative diagram of differential schemes,

Diff(GL , ∂A ) ×C G / Diff(GL , ∂ ) .
A
π2

Diff(GK , ∂A ) ×C G / Diff(GK , ∂ )
A
Let s be a Kolchin closed point of Diff(GK , ∂A ). The projection π2 of the above
diagram is exhaustive. Consider any p ∈ π2−1 (s), and let us consider a Kolchin
closed point x in the closure {p}. Thus, π2 (x) is in the closure {s}. As s is a
Kolchin closed point we know that π2 (x) = s. Hence, there is a Kolchin closed
point x ∈ Diff(GL , ∂A ) such that π2 (x) = s.
Consider two Kolchin closed points s, y ∈ Diff(GK , ∂A ). Because of the above
argument there are two Kolchin closed points x, y ∈ Diff(GL , ∂A ) such that π2 (x) =
s and π2 (y) = y. The set of Kolchin closed points of Diff(GL , ∂A ) is a G(C)-
homogeneous space in the set theoretical sense. Then there is σ ∈ G(C) such that
x · σ = y, and by the commutativity of the diagram we have s · σ = y.
Corollary 4.27. Let x and y be two closed points of Diff(GK , ∂A ). Then there
exists an invertible K-isomorphism of differential fields κ(x) κ(y).
Proof. There is a closed point σ ∈ G, such that x · σ = y. Then
Rσ : (GK , ∂A ) → (GK , ∂A )
is an automorphism that maps x to y. Then it induces an invertible K-isomorphism
Rσ : κ(y) → κ(x).

Definition 4.28. For each closed point x ∈ Diff(GK , ∂A ) we say that the differ-
ential extension K ⊂ κ(x) is a Galois extension associated to the non-autonomous
differential algebraic dynamical system (GK , ∂A ).
Notation. As we have proven, all Galois extensions associated to (GK , ∂A )
are isomorphic. From now on let us choose a closed point x and denote by K ⊂ L
its corresponding Galois extension.
Proposition 4.29. A Galois extension is a minimal splitting extension for
(GK , ∂A ) in the following sense: If K ⊂ S is any splitting extension for (GK , ∂A )
then there is a K-isomorphism of differential fields L → S.
Proof. If K ⊂ S is an splitting extension, then (GS , ∂A ) splits. Hence, for
each Kolchin closed differential point x ∈ Diff(GS , ∂A ) the rational field of x is S.
Let us consider the natural projection π : (GS , ∂A ) → (GK , ∂A ). We can choose a
Kolchin closed point x ∈ Diff(GK , ∂A ) such that π(x) = x. We have a morphism of
K-differential algebras between the corresponding rational fields π : L → S.
Example 4.30 (Picard-Vessiot extensions). Let us consider system of n linear
differential equations
∂x = Ax, A ∈ gl(n, K),
and let us denote aij for the matrix elements of A. The algebraic construction of
the Picard-Vessiot extension is done as follows (cf. [19] and [41]):
Let us consider the algebra K[uij , ∆], being ∆ = |uij |−1 the inverse of the
determinant. Note that it is the algebra of regular functions on the affine group
GL(n, K). If is an affine group, and then it is isomorphic to the spectrum
GL(n, K) = Spec(K[uij , ∆]).
We define the following derivation,

n
∂A uij = aik ujk ,
k=1
that gives to K[uij , ∆] the structure of differential K-algebra, and to (GL(n, K), ∂A )
the structure of automorphic system. The set of Kolchin closed differential points
od Diff(GL(n, K), ∂A ) is the set of maximal differential ideals of R. A Picard-Vessiot

algebra is a quotient algebra K ⊂ K[uij , ∆]/m, and a Picard-Vessiot extension is
a rational differential field K ⊂ κ(m). It is self-evident that the Picard-Vessiot
extension is the particular case of Galois extension when the considered group is
the general linear group.
Lemma 4.31. Let K ⊂ S be a splitting extension. The canonical projection
π : Diff(GS , ∂A ) → Diff(GK , ∂A )
is a closed map.
Proof. It is enough to prove that the projection y = π(y) of a closed point

y ∈ Diff(GS , ∂A ) is a closed point. Let us take a closed point z ∈ {y}. Then
π −1 (z) is closed and there is a closed point z ∈ π −1 (z). Diff(GS , ∂A ) is a principal
homogeneous G-space, there is a σ ∈ G(C) such that z · σ = y, and then z · σ = y.
G(C) acts transitively in the space of closed points, and z is closed, so that we have
proven that y is closed. In fact y and z are the same differential point.
Proposition 4.32. Let us consider any intermediate differential extension,

K ⊂ F ⊂ S, with K ⊂ S an splitting extension. The projection,
π : Diff(GF , ∂A ) → Diff(GK , ∂A ),
is a closed map.
Proof. Let us consider the following diagram of projections:
Diff(GS , ∂A )
π1
/ Diff(GK , ∂ )
PPP nn7
A
PPPπ2 πnnnn
PPP nnn
PP' nnn
Diff(GF , ∂A )
By Lemma 4.31 π1 and π2 are closed and surjective. Then π is closed.
Lemma 4.33. Let K ⊂ F ⊂ L be an intermediate differential extension of the

Galois extension of (GK , ∂A ), and σ the fundamental solution associated to x. Let
us consider the sequence of base changes,
Diff(GL , ∂A )
π1
/ Diff(GF , ∂ ) π2
/ Diff(GK , ∂ )
A A
σ / y / x ,
then y is closed in Kolchin topology, κ(y) is the Galois extension L and σ is the
fundamental solution associated with y.
Proof. By Proposition 4.31 π1 is a closed map, so that y is a closed point. The

chain of projections induces a chain of differential extensions κ(x) ⊆ κ(y) ⊆ κ(σ)
but κ(x) = κ(σ), and then we have the equality.
4.9. Galois Group. Here we give a purely geometrical definition for the Ga-
lois group associated to a Kolchin closed differential point. We prove strong nor-
mality of the Galois extensions, and identify our geometrically-defined Galois group
with the group of automorphisms of the Galois extension. Let us consider the action
of G on Diff(GK , ∂A ) shown in Subsection 4.7:
Diff(GK , ∂A ) ×C G → Diff(GK , ∂A ).
Definition 4.34. Let x ∈ Diff(GK , ∂A ) be a Kolchin closed differential point.

We call Galois group of the system (GK , ∂A ) in x to the isotropy subgroup of x in
G by the above action, and denote it by Galx (GK , ∂A ).
Proposition 4.35. Galx (GK , ∂A ) is an algebraic subgroup of G.
Proof. Denote by Hx the Galois group in x. Let us consider the projection π1

from GK to G induced by the extension C ⊂ K. Denote by x the point π1 (x), and
let U be an affine neighborhood of x. Then U = G \ Y with Y closed in G.
UK is an affine neighborhood of x in GK . We have that the ring of regular
functions in UK is the tensor product OG (U ) ⊗C K. We identify x with a maximal
prime differential ideal x ⊂ OG (U )⊗C K. Let us consider a C-point σ of G. Then, for
each f ∈ OG (U )⊗C K we have that the right translate Rσ (f ) is in OG (U ·σ −1 )⊗C K.
The morphism
π2 : G → G, σ → Rσ (x),
is algebraic, and let W be the complementary in G of π2−1 (Y ),
W = G \ π2−1 (Y ),
W is an open subset in G verifying:

(a) for all σ ∈ W (C), x ∈ U ∩ U · σ −1 ,
(b) Hx ⊂ W .
We will prove that the equations of Hx in W are algebraic. Let us consider W1 an
affine open subset in W . Let {ξ1 , . . . , ξr } be a system of generators of OG (W ) as
C-algebra. The composition is algebraic,
π3 : U ×C W1 → G, (y, σ) → y · σ,
and it induces a morphism,
π3 : OG,x → (OG (U ) ⊗C O(W1 ))π−1 (x) ,

3
and then for each f ∈ OG,x , π3 (f ) = F (ξ), is a rational function in the ξi with
coefficients in OG,x . We identify x with a prime ideal of OG (U ) ⊗C K. We consider
a system of generators,
x = (η1 , . . . , ηr ), ηi ∈ OG (U ) ⊗C K.
Property (b) says that by the natural inclusion,
j : OG (U ) ⊗C K → (OG (U ) ⊗C O(W1 ))π−1 (x) ⊗C K,

3
j(x) spans a non trivial ideal of (OG (U ) ⊗C O(W1 ))π−1 (x) ⊗C K, and then we have
3
a commutative diagram:
OG (U ) ⊗C K / (OG (U ) ⊗C O(W1 ))π−1 (x) ⊗C K .
3
π4

κ(x) / (κ(x) ⊗C O(W1 ))π−1 (x)
3
An element σ ∈ W1 stabilizes x if and only if Rσ (ηi ) ∈ x, and this is so if and only
if π4 (j(ηi )) = 0 for i = 1, . . . , r. Let us consider a basis {eλ }λ∈Λ of κ(x) over C. For
each i, we have a finite sum:

Giα (ξ)eα
π4 (j(ηi )) = α ,
β H iβ (ξ)eβ
and then Giα (ξ) ∈ O(W1 ) are the algebraic equations of Hx in W1 .

Remark 4.36. Let x be a Kolchin closed differential point as above, and H ⊂ G
the Galois group of (GK , ∂A ) in x. Then HK = H ×C Spec(K) is the stabilizer
subgroup of {x}, the Zariski closure of x, by the action of composition by the right
side:
GK ×K GK → GK .
However, the morphisms Rσ for σ ∈ HK are not in general morphisms of schemes
with derivation. In the same sense, for any field extension K ⊂ L, HL ⊂ GL is the
stabilizer group of π −1 (x), the Zariski closure of the preimage of x, where π is the
natural projection from GL to GK . This means that HL stabilizes the fiber, in the
following sense: for each L-point σ ∈ HL , Rσ : GL → GL induces,
Rσ |π−1 (x) : π −1 (x) → π −1 (x).
Proposition 4.37. Consider two Kolchin closed differential points x, y in Diff(GK , ∂A ).
The groups Galx (GK , ∂A ) and Galy (GK , ∂A ) are isomorphic conjugated algebraic
subgroups of G.
Proof. The group of C-points of G acts transitively in the set of closed differ-
ential points. Hence, there exists σ ∈ G(C) with x·σ = y, and then Hx ·σ = σ·Hy .
Theorem 4.38. The Galois extensions associated to (GK , ∂A ) are strongly nor-
mal extensions.
Proof. Let us consider a Galois extension K ⊂ L. Thus, L is the rational
field of certain Kolchin closed differential point that we denote by x. Let us consider
σ ∈ GL the fundamental solution associated to x. We have that σ projects onto x and
the gauge transformation Lσ−1 is a splitting morphism. We define the morphism
ψ of schemes with derivation trough the following commutative diagram:
(GL , ∂A )
π / (GK , ∂ )
O hhh4
A
hhhh
hhhh
Lσ
hhhh ψ
hhhh
(GL , ∂) = G ×C (Spec(L), ∂)
Denote by H the Galois group in x. We have that (HL , ∂) ⊂ (GL , ∂) is a closed
subscheme with derivation. The group HL is the preimage of H by the projection
from GL to G. By remark 4.36 HL is the stabilizer of the π −1 (x) in GL . It means

that for any point z of GL whose projection is addherent to x and any L-point τ
of HL , the right translate z · τ is also addherent to x. In particular we have that
ψ(τ ) = x, and then
(HL , ∂) ⊂ ψ −1 (x).
Reciprocally, let us consider an L-point τ ∈ ψ −1 (x). Therefore π(σ · τ ) is addherent
to x. The following diagram is commutative:
GL ×L GL / GL

GK ×K GK / GK
We deduce that, for any other preimage σ̄ of x by π, the right translated σ̄ · τ also
projects onto {x}. Thus, τ stabilizes π −1 (x), so that τ ∈ (HL , ∂). Finally we have
the identity:
ψ −1 (x) = (HL , ∂) = H ×C (Spec(L), ∂).
On the other hand we apply the affine stalk formula (Proposition A.4, that
comes from the classical stalk formula, Theorem A.1, in Appendix A) to x. We
obtain the isomorphism:
π −1 (x) (Spec(L ⊗K L), ∂).
From the definition of ψ we know that Lσ gives us an isomorphism between
the fibers π −1 (x) and ψ −1 (x). This restricted morphism Lσ |(HL ,∂) is a splitting
morphism
(Spec(L ⊗K L), ∂)
π /5 {x}
O kkkkk
k
kkk
Lσ |(HL ,∂)
kk kkk ψ
kkk
H ×C (Spec(L), ∂)
of the tensor product L ⊗K L. All differential point τ ∈ (Spec(L ⊗K L, ∂) must be

be in the preimage of x, because of the maximality of x as differential point if GL .
If follows that Diff(Spec(L ⊗K L, ∂) = DiffSpec(L ⊗K L). And then, we obtain an
isomorphism
DiffSpec(L ⊗K L) → H ×C DiffSpec(L),
it follows that K ⊂ L is strongly normal.
Remark 4.39. Following [22], DiffSpec(L ⊗K L) is the set of admissible K-

isomorphism of L, modulo generic specialization. In the case of a strongly normal
extension K ⊂ L the space of constants Const(DiffSpec(L ⊗K L)) is an algebraic
group and its closed points correspond to differential K-algebra automorphisms of
L. Let us consider the previous splitting morphism,
H ×C (Spec(L), ∂) → (Spec(L ⊗K L), ∂)
if we apply the constant functor Const, we obtain a isomorphism of C-algebraic
varieties,
s
H−→ Gal(L/K),
where H and Gal(L/K) are algebraic groups. To each τ ∈ H, we have x · τ = x, and

the Rτ : L → L. We have Rτ ◦ Rτ̄ = Rτ τ̄ and it realizes H as a group of differential
K-algebra automorphisms of L.
Theorem 4.40. The Galois group Galx (GK , ∂A ) is the group of differential
K-algebra automorphisms of the Galois extension K ⊂ κ(x).
Proof. Denote, as above, by H ⊂ G the Galois group and by L the Galois
extension κ(x). We consider the isomorphism s stated in remark 4.39. Let us prove
that s is an isomorphism of algebraic groups over C, and that for τ ∈ H(C), s(τ ) is
the automorphism Rτ of L, induced by the translation Rτ .
We already know that s is a scheme isomorphism. We have to prove that it is a
group morphism. For τ ∈ H, let us compute s(τ ). First, let us denote by τ̄ the point
of HL obtained from τ after the base extension from C to L. It is a differential point
of (HL , ∂). Then Lσ (τ̄ ) = Rτ (σ) ∈ π −1 (x). We identify Rτ (σ) with a differential
point of π −1 (x). By the stalk formula we have that π −1 (x) = (Spec(OGK ,x ⊗K
L), ∂). We identify Rτ (σ) with a prime differential ideal of OGK ,x ⊗K L. Because
π(Rτ (σ)) = x, the morphism Rτ (σ) factorizes,
OGK ,x ⊗K L
QQQ
QQQ R (σ)
QQτQ
x ⊗Id QQQ

QQQ
κ(x) ⊗K L /( L
ψ
and then the kernel of ψ is the prime differential ideal defining the automorphism
s(τ ),
ψ(a ⊗ b) = s(τ )(a) · b
Let us consider the right translation Rτ ,
GL
Rτ
/ GL σ
Rτ
/ Lσ (τ̄ )

GK / GK x /x
we have a commutative diagram between the local rings,
LO o
Id
LO ,
σ Rτ (σ)
Rτ
OGL ,π−1 (x) o OGL ,π−1 (x)
O O
OGK ,x o OGK ,x
Rτ
where OG,π−1 (x) = OGK ,x ⊗K L, and the morphism Rτ on these rings is defined as
follows:
OGK ,x ⊗K L → OGK ,x ⊗K L, a ⊗ b → Rτ (a) · b.
It is then clear that morphism ψ defined above sends,
ψ : (a ⊗ b) → Rτ (a) · b
and then its kernel defines the automorphism Rτ and we finally have found Rτ =
s(τ ).
4.10. Galois Correspondence. There is a Galois correspondence for strongly
normal extensions (theorem 3.23). It is naturally transported to the context of al-
gebraic automorphic systems. Let L be a Galois extension, which is the rational
field κ(x) of a Kolchin closed point x as above. Let F be an intermediate differential
extension,
K ⊂ F ⊂ L.
We make base extensions sequentially so that we obtain a sequence of schemes with
derivations,
(GL , ∂A ) → (GF , ∂A )) → (GK , ∂A ),
and the associated sequence of differential schemes,
Diff(GL , ∂A ) → Diff(GF , ∂A ) → Diff(GK , ∂A ).
Let σ ∈ G(L) be the fundamental solution induced by x. We obtain a sequence of
differential points:
σ → y → x.
They are Kolchin closed and σ is the fundamental solution associated to x and y
(Lemma 4.33). The stabilizer subgroup of y is a subgroup of the stabilizer subgroup
of x. We have inclusions of algebraic groups,
Galy (GF , ∂A ) ⊂ Galx (GK , ∂A ) ⊂ G.
In particular we have that K ⊂ F is a strongly normal extension if and only if
Galy (GF , ∂A ) Galx (GK , ∂A ).
Proposition 4.41. Assume that Galx (GK , ∂A ) is the whole group G, and K ⊂
F is a strongly normal extension. Then the quotient group
Ḡ = G/Galy (GF , ∂A )
be the projection of A
exists. Let B in R(Ḡ) ⊗C K. Then, there is a unique closed
differential point z ∈ Diff(ḠK , ∂B ), and,
Galz (ḠK , ∂B ) = Ḡ.
Proof. The quotient realizes itself as the group of automorphisms of the dif-
ferential K-algebra F. The extension K ⊂ F is strongly normal, and then this group
is algebraic by Galois correspondence (Theorem 3.23). The induced morphism
π : Diff(GK , ∂A ) → Diff(ḠK , ∂A )
=
restricts to the differential points, and it is surjective. The hypothesis Galx (GK , A)
G implies that Diff(GK , ∂A ) consist in the only point {x}, and then Diff(ḠK , ∂A ) =
{z}. Hence, z is the generic point of GK and the Galois group is the total group.
Reciprocally let us consider an algebraic subgroup H ⊂ Galx (GK , ∂A ). Then H
is a subgroup of differential K-algebra automorphisms of L. Let F = LH be its field
of invariants. We have again a sequence of non-autonomous algebraic dynamical
systems
(GL , ∂A ) → (GF , ∂A ) → (GK , ∂A ).
Let again σ be the fundamental solution induced by x, we have the sequence of

closed differential points,
σ → y → x
Proposition 4.42. Let us consider an intermediate differential field,
K ⊂ F ⊂ L,
as above, and H = Aut(L/F), then
(a) H is the Galois group Galy (GF , ∂A ) ⊂ Galx (GK , ∂A ).
(b) K ⊂ F is strongly normal if and only if H Galx (GK , ∂A ). In such case
Aut(F/K) = Galx (GK , ∂A )/H.
Proof. By considering the identification of the Galois group with the group of
automorphisms, the result is a direct translation of the Galois correspondence for
strongly normal extensions (see [22] Theorem 20.5, Theorem 3.23 in this text).
In particular, each algebraic group admits a unique normal subgroup of finite
index, the connected component of the identity. Let Gal0x (GK , ∂A ) be the connected
component of the identity of Galx (GK , ∂A ) and,
Gal1x (GK , ∂A ) = Galx (GK , ∂A )/Gal0x (GK , ∂A ),
which is a finite group. In such case we have:
0
(a) The invariant field LGalx (GK ,∂A ) is the relative algebraic closure K◦ of K
in L.
(b) K ⊂ K◦ is an algebraic Galois extension of Galois group Gal1x (GK , ∂A ).
(c) Galy (GK◦ , ∂A ) = Gal0x (GK , ∂A ).
Thus, we can set out:
Proposition 4.43. K is relatively algebraically closed in L if and only if its
Galois group is connected.
4.11. Galois Correspondence and Group Morphisms. Here, we relate
the Galois correspondence and the projection of automorphic vector fields through
algebraic group morphisms. It is self evident that a group morphism π : G → Ḡ
sends an automorphic system A in G with coefficients in K to an automorphic
system π(A) in Ḡ with coefficients in K. Furthermore we know that π(A)
is an
automorphic system in the image of π which is a subgroup of Ḡ. By restricting our
analysis to this image, we can assume that π is a surjective morphism.
Theorem 4.44. Let π : G → Ḡ be a surjective morphism of algebraic groups,
the projected automorphic system π(A).
and B Then:
(1) y = π(x) is a closed differential point of Diff(ḠK , ∂B ).
(2) κ(y) is a strongly normal intermediate extension of K ⊂ κ(y) ⊂ L.
(3) Galy (ḠK , ∂B ) = Galx (GK , ∂A )/(ker(π) ∩ Galx (GK , ∂A )).
(4) Let z be a Kolchin closed point of (Gκ(y) , ∂A ) in the fiber of x. Then
Galz (Gκ(y) , ∂A ) = ker(π) ∩ Galx (GK , ∂A )
Proof. (1) Let s be a closed point of Diff(ḠK , ∂B ) adherent to y. Then π −1 (x)
is a closed subset of Diff(GK , ∂A ) and it contains a closed point z. G(C) acts
transitively in the set of closed points, and then there is τ ∈ G(C) such as x = z · τ .
Thus, y = s · π(τ ), so that y is closed, s = x, and furthermore π(τ ) ∈ Galy (ḠK , ∂B ).
(2) π : κ(y) → L is a differential K-algebra morphism, and κ(y) is realized as

an intermediate extension K ⊂ κ(y) ⊂ L. It is a strongly normal if and only if
the subgroup of Gal(L/K) fixing κ(y) is a normal subgroup. We identify Gal(L/K)
with Galx (GK , ∂A ). Then τ fixes κ(y) if and only if π(τ ) = e. This subgroup fixing
κ(y) is ker(π) ∩ Galx (GK , ∂A ). By hypothesis, ker(π) is a normal subgroup of G,
and then its intersection with Galx (GK , ∂A ) is a normal subgroup.
Finally, be obtain (3) and (4) by Galois correspondence.
4.12. Lie Extension Structure on Intermediate Fields. Differential field
approach to Lie-Vessiot systems was initiated by K. Nishioka, in terms of the notions
of rational dependence on arbitrary constants and Lie extensions (see definitions
3.24 and 3.25). Here we relate our results with these notions.
Theorem 4.45. Assume one of the following:
(a) K is algebraically closed.
(b) The Galois group of (GK , ∂A ) is G.
Let y be a particular solution of (MK , ∂X ) with coefficients in a differential field
extension K ⊂ R. Assume that R is generated by y. Then:
(i) K ⊂ R depends rationally on arbitrary constants.
(ii) K ⊂ R is a Lie extension.
Proof. (i) R is an intermediate extension of the splitting field of the auto-
morphic system which is a strongly normal extension. It is a stronger condition
than the one of Definition 3.24, thus R depends rationally on arbitrary constants.
(ii) If K is algebraically closed, then the result comes directly from Theorem
3.26. For the case (b), some analysis on the infinitesimal structure of R is must be
done. If the Galois group is G, then there are not non-trivial differential points in
GK , nor in MK . Then R coincides with M(MK ), the field of meromorphic functions
in MK . Fundamental vector fields of the action of G on M induce derivations of
the corresponding fields of meromorphic functions so that we have a Lie algebra
morphism,
R(G) → DerK (R), A i → X
i,
and the derivation in ∂ in R is seen in M(R) as the Lie-Vessiot system

r
∂¯ = ∂ + i.
fi X
i=1
From that, we have that,
¯ R(G)] ⊂ R(G) ⊗C K,
[∂,
and because the vector fields X i span the tangent vector space to M , we have that
the morphism,
R(G) ⊗C R → DerK (R)
is surjective. According to Definition 3.25 we conclude that R is a Lie extension.
5. Algebraic Reduction and Integration

Here we present the algebraic theory of reduction and integration of algebraic
automorphic and Lie-Vessiot systems. Our main tool is an algebraic version of Lie’s
reduction method, that we call Lie-Kolchin reduction. Once we have developed this
tool we explore different applications.
5.1. Lie-Kolchin Reduction Method. In [4], when discussing the general

topic of analytic Lie-Vessiot systems, we have shown the Lie’s method for reducing
an automorphic equation to certain subgroups, once we know certain solution of a
Lie-Vessiot associated system. This method is local, because it is assumed that we
can choose a suitable curve in the group for the application of the algorithm. A
germ of such a curve exists, but it is not true that a suitable global curve exists in
the general case. In the algebraic realm we will find obstructions to the applicability
of this method, highly related to the structure of principal homogeneous spaces over
a non algebraically closed field, and then to Galois cohomology.
We will show that the application of the Lie’s method in the algebraic case
leads us directly to Kolchin reduction theorem of a linear differential system to the
Lie algebra of its Galois group. Because of this, we decided to use the nomenclature
of Lie-Kolchin reduction method.
5.2. Lie-Kolchin Reduction. From now on, let us consider a differential
field K of characteristic zero. The field of constant is C, that we assume to be
algebraically closed. Let G be an algebraic group over C, and let A be an algebraic
automorphic vector field in G with coefficients in K. We also fix a Kolchin closed
point x of Diff(GK , ∂A ) and denote by L its associated Galois extension.
Lemma 5.1. Let G ⊂ G be an algebraic subgroup, and let M be the quotient
homogeneous space G/G . Then:
(a) MK = GK /GK
(b) Let us consider the natural projection morphism πK : GK → MK . For each
−1
rational point x ∈ MK , πK (x) ⊂ GK is an homogeneous space of group

GK .
Proof. (a) C is algebraically closed, and then the geometric quotient is uni-
versal; (a) is the fundamental property of geometric universal quotients (see [34]).
(b) The isotropy subgroup Hx of x is certain algebraic subgroup isomorphic and
−1
conjugated with GK . The action of (Hx )K on G preserves the stalk πK (x),
−1 −1
ψ : (Hx )K ×K πK (x) → πK (x),
the induced morphism
−1 −1 −1
(ψ × Id) : (Hx )K ×K πK (x) → πK (x) ×K πK (x)
is the restriction of the isomorphism
GK ×K GK → GK ×K GK , (τ, σ) → (τ · σ, σ),
and then it is an isomorphism.
Let M be an homogenous space over G, and X the Lie-Vessiot vector field

induced in M by the automorphic vector field A. Let us fix a rational point x0 of
M and denote by Hx0 the isotropy subgroup at x0 .
Lemma 5.2. Assume that x0 ∈ M is a constant solution of (MK , ∂X ). Then:
∈ R(Hx ) ⊗C K.
A 0
Proof. There is a solution τ of A with coefficients in L such that x0 = τ · x0 .

Therefore τ ∈ (Hx0 )L and its logarithmic derivative is an automorphic vector field
in Hx0 ,
l∂(τ ) ∈ R(Hx0 ) ⊗C L.
∈ R(Hx ) ⊗C K.
we obtain A
Taking into account that l∂(τ ) = A,
0
Theorem 5.3 (Main Result). Let us assume that (MK , ∂X ) has a solution x
with coefficients in K. If H 1 (Hx0 , K) is trivial, then there exists a gauge transfor-
mation Lτ of GK that sends the automorphic vector field A to:
= Adjτ (A)
B + l∂(τ ),
∈ R(Hx0 ) ⊗C K an automorphic vector field in Hx0 .

with B
Proof. Let us consider the canonical isomorphism G/Hx0 → M that sends
the class [σ] to σ · x0 . Now, let us consider the base extended morphism,
π : GK → MK , τ → τ · x0 .
We are under the hypothesis of Lemma 5.1 (b). Therefore the stalk π −1 (x) is a
principal homogeneous space of group (HK )x which is a subgroup of GK conjugated
to (Hx0 )K . Because of the vanishing of the Galois cohomology, there exist a rational
point τ1 ∈ π −1 (x), and then τ1 · x0 = x. Define τ = τ1−1 . Let us consider the gauge
transformation,
Lτ : (GK , ∂A ) → (GK , ∂B ) Lτ : (MK , ∂X ) → (MK , ∂Y ),
where Y We have that τ · x = x0

is the Lie-Vessiot vector field in M induced by B.
is a constant solution of (MK , ∂Y ). By Lemma 5.2, B is an automorphic field in
Hx0 .
Proposition 5.4. Assume that there is a rational point x0 ∈ M such that

Galx (GK , ∂A ) ⊂ Hx0 , then there exists a rational solution x ∈ M (K) of X.

Proof. Let us consider the fundamental solution σ associated to x. We con-

sider it as an L-point of G,
σ : Spec(L) → GK .
It is determined by the canonical morphism of taking values in σ,
σ : OGK ,x → L = κ(x).
Now, let us consider the projection π : G → M , τ → τ · x0 . It induces a
morphism π : GK (L) → MK (L). Let us consider x = π(σ). This point x is an L
point of M and then it is a morphism
x : Spec(L) → MK .
Let x̄ ∈ MK be the image of x; then x is determined by the morphism x defined
by the following composition:
OMK ,x̄ S π / OGK ,x

SSSS E
SSSS EEE σ
SSSS EE
SSSSEE
x
S")
L
We are going to prove that x is a rational point of MK . Let us consider
τ ∈ Galx (GK , ∂A ). Therefore we have Rτ (x) = x, and the following diagram is
commutative:
OMK ,x̄ VV
00 II VVVV
00 IIII VVVVVVx
00 II VVVV
I$ VVVV
00 VVV*
00 OGK ,x /L
00 σ
||
x 00 ||
00 | ||
|
00 (στ ) |||
00 || R
||
τ
00 |
}||
L
For each f ∈ OXK ,x̄ , we have x (f ) = Rτ (x (f )). This equality holds for all
τ ∈ Hx0 . Hence, x (f ) an element of L that is invariant for any differential K-
algebra automorphism of L. In virtue of the Galois correspondence the fixed field
of L by the action of Galx (GK , ∂A ) is K . Thus, x (f ) ∈ K.
Theorem 5.5. Let us consider an algebraic subgroup G of G verifying:

(1) Galx (GK , ∂A ) ⊂ G ,
(2) H 1 (H, K) is trivial.
Then there exist a gauge isomorphism Lτ of G with coefficients in K reducing the
automorphic system A to an automorphic system in H,
= Adjτ (A)
B + l∂(τ ),
belongs to R(G ) ⊗C K.
Proof. By Proposition 5.4 there exists a rational solution of the Lie-Vessiot
Theorem 5.3 says that such a reduction exists.
system in M associated to A.
Denote by Gal0x (GK , ∂A ) the connected component of the identity of the Galois
group Galx (GK , ∂A ).
Corollary 5.6. Let K◦ be the relatively algebraic closure of K in L. Assume
that H 1 (Gal0x (GK , ∂A ), K◦ ) is trivial. Then there is a gauge transformation Lτ , τ
with coefficients in K◦ such that
= Adj (A)
B + l∂(τ )
τ
belongs to R(Gal0x (GK , ∂A )) ⊗C K◦ .

Proof. We know that the Galois group of the automorphic system with coef-
ficients in K◦ is precisely Gal0x (GK , ∂A ) (see, for instance, remark (c) in [3], below
Proposition 18). We apply then Theorem 5.5.
Corollary 5.7. If H 1 (Galx (GK , ∂A ), K) is trivial then Galx (GK , ∂A ) is con-
nected.
Proof. If H 1 (Galx (GK , ∂A ), K) is trivial, then we can reduce the automorphic
system to an automorphic system in R(Galx (GK , ∂A ))⊗C K. Note that Gal0x (GK , ∂A )
and Galx (GK , ∂A ) have the same Lie algebra. Therefore the Galois group of the
reduced equation is contained in Gal0x (GK , ∂A ).
The following is an extension of the classical result of Kolchin and Kovacic on

the reduction a system of linear differential equations to the Lie algebra of its Galois
group [19]
Theorem 5.8 (Kolchin). Let us consider the relative algebraic closure K◦ of
K in L. There is a gauge transformation Lτ , τ with coefficients in K◦ , such that,
+ l∂(τ )
= Adjτ (A)
B
belongs to R(Galx (GK , ∂A )) ⊗C K◦ .
Proof. Denote by H the Galois group Galx (GK , ∂A ). Let us consider M =
G/H, and let us denote by x0 ∈ M the origin which is the class of H in M . Let
be the Lie-Vessiot vector field in M associated to A.
Y In virtue of Proposition
5.4, the canonical projection G(L) → M (L) sends the fundamental solution σ to a
solution x of (M, ∂Y ) with coefficients in K. Let us consider the projection:
π : GK → MK .
−1
Lemma 5.1 says that the stalk π (x) is a principal homogeneous space modeled
over the group HK . Let us denote by P ⊂ GK such homogeneous space. Note that
P is {x}, the closure of x in Zariski topology. We have the isomorphism,
ψ : P ×K HK → P ×K P, (τ, g) → (τ, τ g),
Let τ be a closed point of P . Its rational field κ(τ ) is an algebraic extension
of K. We have that x = τ · x0 . Thus, we can apply Lie-Kolchin reduction method.
Lτ −1 is a gauge transformation with coefficients in κ(τ ):
Lτ − 1 : Gκ(τ ) → Gκ(τ ) ,
that sends the automorphic vector field A to an automorphic vector field B in H
with coefficients in κ(τ ).
In order to finish the proof we have to see that κ(τ ) is a subfield of the relative
algebraic closure K◦ of K in L. It is enough to see that K ⊂ κ(τ ) is an intermediate
differential extension of K ⊂ L. Furthermore, if κ(τ ) is an intermediate differential
extension then it coincides with K◦ because of the Galois correspondence.
Let us consider then the following base extension and natural projection,
Pκ(τ ) = P ×K Spec(κ(τ )), π1 : Pκ(τ ) → P.
The product Pκ(τ ) is a principal homogeneous space modeled over Hκ(τ ) . Moreover,
τ induces a rational point of Pκ(τ ) . Hence, the Galois cohomology cohomology class
of Pκ(τ ) is trivial, so that it is isomorphic to Hκ(τ ) as homogeneous space. Pκ(τ )
has as many connected components as Hκ(τ ) . We write it as the disjoint union of
its connected components.
Pκ(τ ) = Pi .
i∈Λ
For each i ∈ Λ, the restriction Pi → P is an isomorphism of K-schemes, and π1 is a
trivial covering. But each Pi is a κ(τ )-scheme, and then each component induces in
P an structure of κ(τ )-scheme. Hence we have a realization of κ(τ ) as intermediate
extension
K ⊂ κ(τ ) ⊂ L.
Thus, κ(τ ) = K◦ .
5.3. Integrability by Quadratures. To integrate an automorphic system

by quadratures means to write down a fundamental solution by terms of a formula.
This formula should involve the solutions of certain simpler equations. We assume
that we have a geometrical meccano to express these solutions. We refer to elements
of such a meccano as quadratures. Those simpler equations are like the building
blocks of our integrability theory. Depending of which simpler equations we consider
as integrable we obtain different theories integrability. In theory of Lie-Vessiot
systems the elements of our formulas are the exponential maps of Lie groups and
indefinite integrals.
From a geometric point of view, it is reasonable to consider automorphic sys-
tems in abelian groups as integrable. Let us consider an abelian Lie group G. Then,
the exponential map,
exp : R(G) → G,
is a group morphism, and moreover, R(G) is the universal covering of G. An
automorphic equation,
d log n
(x) = fi (t)A i ∈ R(G)
i, A
dt i=1
is integrated by the formula,
n
t
σ(t) = exp i
fi (ξ)dξ A .
i=1 t0
This formula involves the integral of t dependent functions, and the exponential
map of the Lie group. Assuming that we are able of realize these operations a
reasonable point of view is to consider al automorphic equations in abelian groups
integrable. This assumption is done in [43], and followed in [6]. On the other
hand, the algebraic case has a new kind of richness. An abelian Lie group splits in
direct product of circles an lines, but an abelian algebraic group can carry a higher
complexity, for example in the case of abelian varieties. In such case the exponential
map is the solution of the Abel-Jacobi inversion problem. In [18] Kolchin develops a
theory of integrability generalizing Liouville integrability, in which just quadratures
in one dimensional abelian groups are allowed. It reduces the case to quadratures
in the additive group, the multiplicative group and elliptic curves.
5.4. Quadratures in the Additive Group. Let us consider an automorphic
equation in the additive group C. The additive group is its own Lie algebra, and
the logarithmic derivative is the usual derivative. Thus, the automorphic equations
are written in the following form:
(5.1) ∂x = a, a ∈ K.
Definition 5.9. An extension of differential fields K ⊂ L is an integral exten-
sion if L is K(b), with ∂b ∈ K. We say that b is an integral element over K.
It is obviousthat the Galois extension of equation (5.1) is an integral extension
of K, with b = a. The additive group (of a field of characteristic zero) has no
algebraic subgroups. Therefore, if a is algebraic over K, then a ∈ K. Hence we have
two different possibilities for integral extensions:
• b ∈ K, Gal(L/K) = {e},
• b ∈ K, Gal(L/K) = C.
5.5. Quadratures in the Multiplicative Group. Let us consider now an

automorphic equation in the multiplicative group. For the complex numbers C∗ the
exponential map is the usual exponential. In the general case of an algebraically
closed field of characteristic zero, we can build the exponential map for C ∗ . However,
it does not take values in C ∗ but in a bigger group. We avoid such a construction,
and then we consider the exponential just as an algebraic symbol. The logarithmic
derivative in C ∗ coincides with the classical notion of logarithmic derivative,
∂x
K∗ → K, x → .
x
The general automorphic equation in the multiplicative group is written as follows:
∂x
(5.2) = a, a ∈ K.
x
Definition 5.10. An extension of differential fields K ⊂ L is an exponential
extension if L = K(b), with ∂b
b ∈ K. We say that b is an exponential element over
K.
C ∗ has cyclic finite subgroups. Then, we can obtain exponential extensions that
are algebraic. There appears the following casuistic:
• Gal(L/K) is the multiplicative group C ∗ if b is transcendent over K.
• Gal(L/K) is a cyclic group (Zn )∗ if bn ∈ K for certain n. It means that
there is c ∈ K that nc
∂c
= a. In such case, bn = c.
Reciprocally, any algebraic Galois extension of K with a cyclic Galois group
is an exponential extension. Here, it is a an essential point that C is algebraically
closed.
5.6. Quadratures in Abelian Varieties. Abelian varieties provide us ex-

amples of non linearizable automorphic systems. For the following discussion, let
us assume that the constant field of K is the field of complex numbers C. Let G be
a complex abelian variety of complex dimension g. Let us consider a basis of holo-
morphic differentials ω1 , . . . , ωg , and A1 , . . . , Ag ,B1 , . . . , Bg a basis of the homology
of G, we can assume that Ai ωj = δij . Define the Jacobi-Abel map,
p p
∼
G− → Cg /Λ, p → ω1 , . . . ωg .
e e
The exponential map is given by the exponential universal covering of the torus
and the inversion of the Jacobi-Abel map.
Cg E
EE
EE
exp EE
E"
G
j
/ Cg /Λ
A projective immersion of G in P(C, d), for d big enough, is given by terms

of theta functions, z → (θ0 (z) : . . . : θd (z)). Hence there are some homogeneous
polynomial constrains {P (θ0 , . . . , θd ) = 0}. The quotient θθji defines a meromorphic
abelian function in G (see [28] Chapter 1, Section 3, p. 30). Let us consider affine
coordinates in G, xi = θi
θ0 . We can project the vector fields of R(Cg ) to G,
∂ ∂
∂θj
∂zi θ0 − ∂θ0
∂zi θj
→ Fij (x1 , . . . , xd ) , Fij (x1 , . . . , xd ) =
∂zi j
∂xj θ02
being Fij abelian functions, and then rational functions in the xj . The automorphic
system in Cg
∂
ai , ai ∈ K
i
∂z i
is seen in A as a non linear system an A,

(5.3) ẋj = ai Fij (x1 , . . . , xd ), {P (1, x1 , . . . xd ) = 0}.
i
If b1 , . . . , bd are integral elements over K such that ∂bi = ai , then the solution of
the automorphic system (5.3) is:
θj (b)
xj = , (θ0 (b) : . . . : θd (b)) .
θ0 (b)
Definition 5.11. A strongly normal extension K ⊂ L whose Galois group is
an abelian variety is called an abelian extension.
For an automorphic system in an abelian variety A we have that the Galois
group is an algebraic subgroup of A. Then its identity component is an abelian
variety. The Galois extension is then,
K ⊂ K◦ ⊂ L,
being K◦ ⊂ L an abelian extension.
Example 5.12. Let us consider an algebraically completely integrable hamil-
tonian system in the sense of Adler, Van Moerbecke and Vanhaecke (see [1])
{H, H2 , . . . , Hn } in C2n . Assume that {Hi (x, y) = hi } are the equations of the
affine part of an abelian variety G. The Hamilton equations,
∂H ∂H
(5.4) ẋi = , ẏi = − , Hi (x, y) = hi
∂yi ∂xi
are an automorphic system H in G with constant coefficients K = C. In the generic
case, G is a non-resonant torus, and then it is densely filled by a solution curve of
the equations (5.4). We conclude that (G, ∂H ) has not proper differential points:
its differential spectrum consist only of the generic point. In such case, the Galois
extension of the system is C ⊂ M(G), the field of meromorphic functions in G.
Example 5.13. Automorphic systems in elliptic curves: Let us examine the
case of an elliptic curve E over C. Assume that E is given as a projective subvariety
of P(2, C) in Weierstrass normal form.
t0 t22 = 4t31 − g2 t20 t1 − g3 t30
We take affine coordinates x = t1
t0
and y = t2
t0
. The Lie algebra R(E) is then
generated by the vector field,
∂ ∂
v = y + (12x2 − g2 )
∂x ∂y
Every automorphic vector field in E with coefficients in K is written in the form

av with a ∈ K. A solution of the automorphic equation is a point of E with values
in the Galois extension L. Such solution have homogeneous coordinates (1 : ξ : η)
such that η = a−1 ∂ξ, and ξ is a solution of the single differential equation,
(5.5) (∂ξ)2 = a2 (4ξ 2 − g1 ξ − g2 ).
If we know a particular solution b of (5.5) then we can write down the general
solution (1 : ξ : η) of the automorphic equation by means of the addition law in E
(see [18] p. 804 eq. 9), depending of an arbitrary point (1 : x0 : y0 ) ∈ E(C):
Sol(5.5) × E(C) → E(L), (b, (1 : x0 : y0 )) → (1 : ξ : η)
2
1 ∂b − ay0
(5.6) ξ(x0 , y0 ) = −b − x0 −
4 a(b − x0 )
3
∂b + ay0 6 ∂b − ay0 1 ∂b − ay0
(5.7) η(x0 , y0 ) = − + (b + x0 ) − .
2a 2 a(b − x0 ) 4 a(b − x0 )
Definition 5.14. Let K ⊂ L a differential field extension. We say that b ∈ L
is a Weierstrassian element if there exist a ∈ K, and g1 , g2 ∈ C, with the polynomial
4x3 − g1 x − g2 having simple roots and such that, (∂b)2 = a2 (4b2 − g1 x − g2 ). The
differential extension K ⊂ K(b, ∂b) is called an elliptic extension.
The Galois extension of the automorphic equation (5.5) is an elliptic extension
of K. It can be transcendent or algebraic. If it is transcendent then its Galois group
is the elliptic curve E, if it is algebraic then its Galois group is a finite subgroup of
E.
Remark 5.15. Let us examine the case of complex numbers: assume that the
field of constants of K is C. The solution of Weierstrass equation is the elliptic
function ℘, and it gives rise to the universal covering of E,
π : C → E, z → (1 : ℘(z) : ℘ (z)).
The automorphic vector field av in E is the projection of the automorphic vec-
∂
tor field a ∂z in C. The solution of the equation in the additive group is given
by an integral element a. Then the a solution of the projected system in E is
(1 : ℘( a) : ℘ ( a)). Then b = ℘( a) is the Weierstrass element of the Galois ex-
tension. Formulas (5.6) and (5.7) are the addition formulas for the Weierstrass ℘
and ℘ functions.
Example 5.16. We obtain the previous situation in the case of one degree of
freedom, algebraic complete integrable hamiltonian systems. Let us consider the
pendulum equation:

ẋ = y y2
(5.8) − cos(x) = h
ẏ = sin(x) 2
It is written as a simple ordinary differential equation depending of the energy
parameter h,
2
dx
= 2h + 2 cos(x),
dt
by setting z = eix , we obtain the algebraic form of such equation, which is an

automorphic equation in an elliptic curve for all values of h except for h = ±1;
2
dz
= −z 3 − 2hz 2 − 1.
dt
The Weierstrass normal form is attained by setting u = −z
4 − 6 h;
1
2 3
du h2 h 1
= 4u3 − u − + .
dt 3 27 16
Hence, the general solution is written in terms of the ℘ functions of invariants
2 3
g2 = h3 and g3 = h27 + 16
1
, for h = ±1:

2 4h + 3πi
z(t) = −4℘(t + t0 ) − h ; x(t) = log −4℘(t − t0 ) − .
3 6
5.7. Liouville and Kolchin Integrability.
Definition 5.17. Let K ⊂ F a differential field extension. Let us break it up
into a tower of differential fields:
K = F0 ⊂ F1 ⊂ . . . ⊂ Fd = L.
We say that K ⊂ F is . . .
(1) . . . a Liouvillian extension if the differential fields Fi can be chosen in such
way that Fi ⊂ Fi+1 is an algebraic, exponential or integral extension.
(2) . . . a strict-Liouvillian extension if the differential fields Fi can be chosen
in such way that Fi ⊂ Fi+1 is an exponential or integral extension.
(3) . . . a Kolchin extension the differential fields Fi can be chosen in such way
that Li ⊂ Fi+1 is algebraic, elliptic, exponential or integral extension.
Liouvillian and strict-Liouvillian extensions are Picard-Vessiot extensions. An
elliptic curve can not be a subquotient of an affine group. Hence, if K ⊂ F is a
Kolchin extension and Gal(F/K) is an affine group, then it is a Liouville extension.
From this perspective, the following classical result is almost self evident:
Theorem 5.18 (Drach-Kolchin). Let K be a field of meromorphic functions of
the complex plane C. Assume that the Weierstrass’s ℘ function is not algebraic over
K. Then ℘ is not the solution of any linear differential equation with coefficients
in K.
Proof. Let us assume that this equation exist, and let K ⊂ F na associated
its Galois extension. Its Galois group Gal(F/K) is an affine group. We have an
intermediate extension:
K ⊂ K(℘, ℘ ) ⊂ F,
This intermediate extension K ⊂ K(℘, ℘ ) is strongly normal and its Galois group
is an elliptic curve. Thus, there is a normal subgroup H Gal(F/K) and an exact
sequence,
0 → H → Gal(F/K) → E → 0
but the quotient group of an affine group is an affine group, and then E is affine.
From the Galois correspondence and some elemental properties of algebraic
groups we also have immediately the characterization of Liouvillian and Kolchin
extensions in terms of their Galois groups.
Proposition 5.19. Let K ⊂ L be a strongly normal extension.

(1) K ⊂ L is a Kolchin extension if and only if there is a sequence of normal
subgroups in Gal(L/K),
H0 H1 . . . Hn = Gal(L/K),
such that dimC Hi /Hi+1 ≤ 1.
(2) K ⊂ L is a strict-Liouville extension if and only if Gal(L/K) is an affine
solvable group.
(3) K ⊂ L is a Liouvillian extension if and only if the identity component
Gal0 (L/K) is a linear solvable group.
Proof. For (1) and (3) see [18]. Let us proof that linear solvable Galois
group implies strict Liouville. Let us consider a resolution of the Galois group
H0 . . . Hn such that each quotient Hi+1 /Hi is a cyclic group, a multiplicative
group or an additive group. This resolution exist by means of Lie-Kolchin theorem.
This resolution split the extension K ⊂ L in a tower of differential fields
Kn ⊂ Kn−1 ⊂ . . . ⊂K0 .,
Each differential extension of the tower is an exponential, integral or algebraic
extension with cyclic Galois group. But an algebraic extension with cyclic group is
a radical extension. The field√C is algebraically closed, hence such radical extension
is generated by the radical n a of a non-constant element of a, and then it is the
Picard-Vessiot extension of the equation,
∂a
∂x = x,
na
which is an exponential extension.
5.8. Integration by Quadratures in Solvable Groups. Let us remind
that along this chapter we are considering an automorphic vector field A with
coefficients in K in an algebraic group G defined over C. We also consider a Kolchin
closed differential point x ∈ Diff(GK , ∂A ) and the associated Galois extension K ⊂ L.
We are going to explain the classical integration by quadratures in terms of Lie-
Kolchin reduction method and Galois correspondence.
Let us consider a normal subgroup H G, and the quotient group Ḡ = G/H.
Let y be the projection in ḠK of x. In virtue of Theorem 4.44 we know that,
K ⊂ κ(y) ⊂ L,
is an intermediate strongly normal extension. Furthermore, the Galois group in y
of the automorphic system with coefficients if κ(y) is the intersection of the Galois
group Galx (GK , ∂A ) with H.
Theorem 5.20. Assume that there is a resolution of G,
H0 H1 . . . Hn = G,
such that dimC Hi /Hi+1 = 1, then K ⊂ L is a Kolchin extension.
Proof. Let us consider the quotients Ḡi = Hn−i+1 /Hn−i . They are algebraic
groups of dimension one. Each Gi is isomorphic to one of the following: the additive
group, the multiplicative group, or an elliptic curve. Each one corresponds to
an integral, exponential, or Weierstrassian quadrature. We prove the theorem by
induction in the length of the resolution. Let us consider the projection π : G →
G/Hn−1 . Define y = π(x) and let K1 be the relative algebraic closure of κ(x) in L.
Then K ⊂ κ(y) is an integral, exponential or elliptic extension and κ(y) ⊂ K1 is an
algebraic extension. Hence, K ⊂ K1 is a Kolchin extension.
Let z be a closed differential point of (GK1 , ∂A ) in the fiber of x. By The-
orem 4.44 Galz (GK1 , ∂A ) ⊂ Hn−1 , and then by Theorem 5.8 there is a gauge
transformation Lτ with coefficients in K1 reducing the automorphic field to an
automorphic field in Hn−1 . Any Galois extension associated to this last equation
is K1 -isomorphic to L. By the induction hypothesis the extension K1 ⊂ L is a
Kolchin extension, hence K ⊂ L is a Kolchin extension.
Theorem 5.21. Assume that G is affine and solvable. Then K ⊂ L is a strict-
Liouville extension.
Proof. The Galois group is a subgroup of G, and then it is a solvable group.
The result comes from Proposition 5.19 (2) together with Theorem 5.20.
Proposition 5.22. If there is a connected affine solvable group H ⊂ G such
that Galx (GK , ∂A ) ⊂ H, then K ⊂ L is a strict-Liouville extension.
Proof. H is connected affine solvable an then it has trivial Galois cohomology.
We can reduce to the group H by means of theorem 5.5. Hence, we are in the
hypothesis of theorem 5.21.
5.9. Linearization. There exist non-linear non-linearizable algebraic groups.
An algebraic group that does not admit any linear representation is called quasi-
abelian. In other words, a quasi-abelian variety is an algebraic group G such that
OG (G) = C. Algebraic groups over an algebraic closed base field C, which are
complete and connected, are called abelian varieties. Since they are complete va-
rieties, they do not admit non-constant global regular functions and then they are
quasi-abelian.
The following results give us the structure of the algebraic groups by terms of
linear and quasi-abelian algebraic groups. See, for instance [34].
Theorem 5.23 (Rayleigh decomposition). Let G be an algebraic group. There
is a unique subgroup X ∈ G such that, X is quasi-abelian and G/X is an affine
group.
Theorem 5.24 (Chevalley-Barsotti-Sancho). Let G be a connected algebraic
group over C, with C an algebraically closed field of characteristic zero. Then there
is a unique normal affine subgroup N ⊂ G such that the quotient G/N is an abelian
variety.
5.10. Reduction by means Chevalley-Barsotti-Sancho Theorem. In
virtue of Chevalley-Barsotti-Sancho theorem (5.24 in appendix B), there is a unique
linear normal connected algebraic group N G such that the quotient G/N and
is an abelian variety V . Let us consider the projection π : G → V . Let B be the

projected automorphic system π(A) in V , and denote by y the image of x by π. We
state the following:
Theorem 5.25. Let M be the field of meromorphic functions in VK . Assume
that Galy (VK , ∂B ) = V , and one of the following hypothesis:
(1) H 1 (N, M) is trivial.
(2) K is relatively algebraically closed in L.
Then, there is a gauge transformation of G with coefficients in M reducing the

automorphic system A to N .
as an automorphic vector field in G with coefficients

Proof. Let us consider A
in M. By Galois correspondence we have:
Gal(L/M) Galx (GK , ∂A ) ∩ N.
If hypothesis (1) holds, then the statement is a particular case of Theorem 5.5.
Let us prove the result in the case of hypothesis (2). By Theorem 5.8 there exists
a gauge transformations whose coefficients are algebraic over M. By hypothesis
Galx (GK , ∂A ) is connected. This group Galx (GK , ∂A ) realizes itself as a principal
bundle over V whose structural group os Gal(L/M). It implies that Gal(L/M) is
also connected. So that M is relatively algebraically closed in L. The coefficients
of the considered gauge transformation are in M, as we wanted to prove.
5.11. Linearization by means of Adjoint Representation. We consider

GL(R(G)) the group of C-linear automorphisms of the Lie algebra R. It is an
algebraic group over C. The adjoint representation
Adj : G → GL(R(G))
is a morphism of algebraic groups. It gives us a linearization of the equations. Let

us consider the center Z(G) and the exact sequence:
0 → Z(G) → G → GL(R(G)) → 0
Denote by B the projection of the automorphic vector field A by the morphism

Adj. It is a linear system and then its Galois extension K ⊂ P is a Picard-Vessiot
intermediate extension of K ⊂ L.
Proposition 5.26. P ⊂ L is a strongly normal extension and Gal(L/P) is an

abelian group.
Proof. The extension P ⊂ L is a Galois extension of A with coefficients in

P, so that it is strongly normal. Its Galois group is, by the Galois correspondence,
the intersection of the Galois group of Galx (GK , ∂A ) with the center Z(G); it is an
abelian group.
5.12. Linearization by means of Global Regular Functions. The ring of

global regular functions Γ(OG , G) is a Hopf algebra, and then it spectrum is a linear
algebraic group L = Spec(Γ(OG , G)). The kernel C of the canonical morphism
π : G → L is, by definition a quasi-abelian variety (see [34]). Let us consider the
exact sequence:
0 → C → G → L → 0.
We proceed as we did in Proposition 5.26, and then we obtain the following result.
Proposition 5.27. Let K ⊂ P be the Picard-Vessiot extension of the auto-

in L. Then P ⊂ L is a strongly normal extension, and the
morphic system π(A)
connected component of the identity of its Galois group is a quasi-abelian variety.
6. Integrability of Linear Equations

This section is devoted to the Liouville integrability of linear differential equa-
tions. Since the development of Picard-Vessiot system it is a rich field of research,
let us cite some important specialized literature [20], [37], [38], [39], [12], [13].
Here, we adopt a slightly different point of view on linear differential equations.
We see them as automorphic systems. It gives us some insight into the geomet-
ric mechanisms that allows quadratures. In this way we are able to measure the
solvability of the Galois groups, in terms of equations in flag varieties and grass-
manians (Theorem 6.2). They are the natural geometrical generalization of Riccati
equations.
From now on let G be a linear connected algebraic group over C. We consider
A an automorphic vector field in G with coefficients in K.

6.1. Flag Variety. We call Borel subgroup of G to any maximal connected

solvable group of G. Borel subgroups are all conjugated and isomorphic subgroups.
The quotient space G/B is a complete variety (see [34] p. 163, th. 10.2).
Definition 6.1. We call flag variety of G to the homogeneous space quotient
G/B, being B a Borel subgroup of G.
The flag variety of G is defined up to isomorphism of G-homogeneous spaces.
Let us consider F lag(G) a flag variety of G, and let (F lag(G), ∂F ) be the induced
Lie-Vessiot system.
Let us see a natural generalization of the well-known theorem of J. Liouville
that relates the integrability by Liouvillian functions of the second order linear
homogeneous differential equation with the existence of an algebraic solution of an
associated Riccati equation. This classical result is the particular case of GL(2, C)
in the following general Liouville’s theorem.
Theorem 6.2. The Galois extension K ⊂ L is Liouvillian if and only if the
flag Lie-Vessiot system (F lag(G), ∂F ) has an algebraic solution with coefficients in
K◦ , the algebraic relative closure of K in L.
Proof. By the Galois correspondence we have that the Galois group of (GK◦ , ∂A )
is the connected identity component of the Galois group of (GK , ∂A ). Assume that
(F lag(G), ∂F ) has an algebraic solution x ∈ F lag(G)(K◦ ). We are under the hy-
pothesis of Theorem 5.8. There is a gauge transformation of GK0 that send A to
an automorphic vector field B in the Borel subgroup B. Then the Galois group
of B with coefficients in K0 is contained in a Borel subgroup. Then the connected
component of Galx (GK , ∂A ) is solvable.
Reciprocally, let us assume that K ⊂ L is a Liouvillian extension. In such
case the identity connected component of the Galois group is contained in a Borel
subgroup B. By Proposition 5.4 there is a solution with coefficients in K◦ of F .
6.2. Automorphic Equations in the General Linear Group.
6.3. Grassmanians. Let us consider E as n-dimensional vector space. Along
this text m-plane will mean m-dimensional linear subspace. For all m ≤ n the
linear group GL(E) acts transitively in the set of m-planes. For an m-plane Em ,
the stabilizer subgroup is an algebraic group, and then the set of m-planes define
an algebraic homogeneous space.
Definition 6.3. We call grassmanian of m-planes of E, Gr(E, m), to the ho-

mogeneous space whose closed points are the m-planes of E. Denote Gr(C, n, m)
the grassmanian of m-planes of C n .
Example 6.4. Gr(C, n, 1) is the space of lines in C n , and then if its the pro-
jective space of dimension n − 1, P(n − 1, C). The Gr(C, n, n − 1) is the space os
hyperplanes and then it is the dual projective space P(n − 1, C)∗ .
In general, m-planes of E are in one-to-one correspondence with (n − m)-planes

of the dual space E ∗ , and then we have the projective duality
Gr(E, m) Gr(E ∗ , n − m).
The action of GL(E) on Gr(E, m) is not faithful. Each scalar matrix of the center
of GL(C, n) fix all m-planes. Thus, the non faithful action of GL(E) is reduced to
a faithful action of the projective group P GL(E).
All grassmanian are projective varieties. There is a canonical embedding of
Gr(E, m) into the projective space of dimension (m n
) − 1, called the plücker embed-
ding:
Gr(E, m) → P(E ∧n ), e1 , . . . , em → e1 ∧ e1 ∧ . . . ∧ em .
For computation in the grassmanian spaces we will use plückerian coordinates.

This system of coordinates is subordinated to a basis in E. Thus, let us consider
a basis {e1 , . . . , en }. Let E1 = e1 , . . . , em be the m plane spanned by the first
m elements of the basis, and define E2 = em+1 , . . . en its complementary. Let
us consider the projection π : E → E2 of kernel E1 . We define the open subset
U ⊂ Gr(E, m),
U = {F : F ⊕ E2 = E}.
For F ∈ U the splitting of the space induces an isomorphism iF : E1 → F . We have

an isomorphism
∼
U−
→ HomC (E1 , E2 ), F → π ◦ iF .
We define the plükerian coordinates of F as the matrix elements of π ◦ iF in the

above mentioned basis. By permuting the elements of the basis we construct a
covering of Gr(E, m) by (m n
) affine open subsets isomorphic to C n(n−m) .
Let us compute plückerian coordinates in Gr(C, m, n) related to the canonical
basis. Let us consider F ∈ Gr(C, m, n), and a basis of F , {x1 , . . . , xm }, xi =
(x1i , . . . , xni ). The matrix,
⎛ ⎞
x11 . . . x1m
⎜ x21 . . . x2m ⎟
⎜ ⎟
⎜ .. .. .. ⎟
⎝ . . . ⎠
xn1 . . . xnm
is of maximal rank. Thus, there is a non vanishing minor of rank m. In particular,

F is in the open subset U if and olny if the minor corresponding to the first m rows
(m)
does not vanish. In such case we define the numbers λij
⎛ ⎞
1 ... 0
⎛ ⎞ ⎜ . .. .. ⎟
x11 . . . x1m ⎛ ⎞−1 ⎜ .. . . ⎟
x . . . x ⎜ ⎟
⎜ x21 . . . x2m ⎟ 11 1m ⎜ 0 ... 1 ⎟
⎜
⎜ .. . .
⎟ ⎜ ..
⎟ ⎝ .
..
. .. ⎠ = ⎜
. ⎟
⎜ (m) (m)
⎟
⎟
⎝ . .. .. ⎠ ⎜ λ11 ... λ1m ⎟
xm1 . . . xmm ⎜ . ⎟
xn1 . . . xnm ⎜ . .. .. ⎟
⎝ . . . ⎠
(m) (m)
λn−m,1 . . . λn−m,m
that are the plückerian coordinates of Em ∈ Gr(C, m, n) in the open affine subset
U related to the split of C n as E1 ⊗ E2 .
6.4. Flag Variety of the General Linear Group. A flag of subspaces of
C n , is a sequence,
E1 ⊂ E2 ⊂ . . . ⊂ En−1 , dimC Ei = i
of linear subspaces of C n . The space F lag(C, n) of flags of C n is an homogeneous
space of GL(C, n), and it is faithful for the action of P GL(C, n). There is a canonical
morphism,

n−1
F lag(C, n) → Gr(C, n, m), E1 ⊂ E2 ⊂ En−1 → (E1 , . . . , En−1 ).
m=1
By Lie-Kolchin theorem the isotropy subgroup of a flag is also a Borel subgroup.
Then, we can state F lag(C, n) is the flag variety of the general linear group. Let us
introduce a system of coordinates in F lag(C, n). Let us consider {e1 , . . . , en } the
canonical basis of C n . Each σ ∈ GL(C, n) defines a flag F (σ) as follows:
σ(e1 ) ⊂ σ(e1 ), σ(e2 ) ⊂ . . . ⊂ σ(e1 ), . . . , σ(en−1 ).
There is a canonical flag corresponding to the identity element. Its isotropy
group is precisely T (C, n) the group of upper triangular matrices. Then two matrices
A, B ∈ GL(C, n) define the same flag if and only if A = BU for certain U ∈ T (C, n).
Then let us consider the affine subset of GL(C, n) of matrices with non vanishing
principal minors. For such a matrix there exist a unique LU decomposition such
that U ∈ T (C, n) and is a lower triangular matrix as follows,
⎛ ⎞
1 0 ... 0
⎜ λ21 1 . . . 0⎟
⎜ ⎟
A=⎜ . . .. .⎟ U
⎝ .. .. . .. ⎠
λn1 λn2 ... 1
Hence the matrix elements λi define a system of affine coordinates in F lag(C, n),
in certain affine open subset. We construct an open covering of the flag space by
permutating the vectors of the canonical base. The canonical morphism

F lag(C, n) → Gr(C, m, n)
m
is easily written in plückerian coordinates:

(m)

m
λij = λi+m,j − λi+m,k λkj .
k=1
6.5. Matrix Riccati Equations. Let us consider an homogeneous linear dif-

ferential equation
ẋ = Ax, A ∈ gl(K, n).
It is seen as an automorphic system that induces Lie-Vessiot systems in each homo-
geneous space. Let us compute the induced Lie-Vessiot systems in the grassmanian
spaces. First, the linear system induces a linear system in (C n )m .
(6.1) Ẋ = AX,
where X is a n × m matrix. We write X = (YU ), being U a m × m matrix and Y a
(n−m)×m matrix. Λm = Y U −1 is the matrix of plückerian coordinates of the space
generated by the m column vectors of the matrix X. Then, Λ̇m = Ẏ U −1 −Λm U̇ U −1 .
If we decompose the matrix A in four submatrices

A11 A12
A=
A21 A22
being A11 of type m × m, A12 of type m × (n − m), A21 of type (n − m) × m, and
A22 if type m × m. Them the matrix linear equation (6.1) splits as a system of
matrix linear differential equations,
U̇ = A11 + A12 Y, Ẏ = A21 U + A22 Y,
from which we obtain the differential equation for affine coordinates in the grass-
manian,
(6.2) Λ̇m = A21 + A22 Λm − Λm A11 − Λm A12 Λm
which is a quadratic system. We call such a system a matrix Riccati equation
associated to the linear system.
⎛ ⎞
(m) (m)
λ11 ... λ1,m
⎜ . ⎟
Λm = ⎜⎝ .
.
..
.
..
.
⎟
⎠
(m) (m)
λn−m,1 . . . λn−m,m
(m)

n−m
(m)

m
(m)
(m) (m)
λ̇ij = am+i,j + am+i,m+k λkj − λik akj − λik ak,r+m λrj
k=1 k=1 k=1...m
r=1...n−m
Example 6.5. Let us compute the matrix Riccati equations associated to the
general linear system of rank 2 and 3. First, let us consider a general linear system
of rank 2,
ẋ1 = a11 x1 + a12 x2 , ẋ2 = a21 x1 + a22 x2 .
There is one only grassmanian Gr(C, 1, 2), which is precisely the projective line.
The associated matrix Riccati equation is an ordinary Riccati equation
ẋ = a21 + (a22 − a11 )x − a12 x2 .
In the case of a general system of rank 3,
⎛ ⎞ ⎛ ⎞⎛ ⎞
ẋ1 a11 a12 a13 x1
⎝ẋ2 ⎠ = ⎝a21 a22 a23 ⎠ ⎝x2 ⎠
ẋ3 a31 a32 a33 x3
there are two grassmanian spaces, Gr(C, 1, 3) and Gr(C, 2, 3), being the projec-
tive plane P2 (C) and the projective dual plane P2 (C)∗ respectively. Then we obtain
two quadratic systems,

ẋ = a21 + (a22 − a11 )x + a23 y − a12 x2 − a13 xy
P(2, C)
ẏ = a31 + (a33 − a11 ) + a32 x − a13 y 2 − a12 xy

ξ˙ = a31 + (a33 − a11 )ξ + a21 η − a23 ξη − a13 ξ 2
P(2, C)∗
η̇ = a32 + (a33 − a22 )η + a12 ξ − a13 ξη − a23 η 2
called the associated projective Riccati equations.
6.6. Flag Equation. From the relation between plückerian coordinates and
affine coordinates in the flag variety we can deduce the equations of the induced
Lie-Vessiot system in F lag(C, n), from the matrix Riccati equations. We will obtain
a Riccati quadratic equation for n = 2, and a cubic system for n ≥ 3.

n
j
j
j
λ̇ij = aij + aik λkj − λik akj + λir λrk akj
k=j+1 k=1 k=1 r=k+1

j

n
j

n
j
− λik akr λrj + λis λsk akr λrj ,
k=1 r=j+1 k=1 r=j+1 s=k+1
Setting λii = 1 for all i, we can simplify these equations.

n
j
n
j
j

n
(6.3) λ˙ij = aik λkj − λik akr λrj + λir λrk aks λsj
k=j k=1 r=j k=1 r=k+1 s=j
Such as cubic system can be seen as a hierarchy of projective Riccati equations.

The equation corresponding to the first column λi1 , i = 2 . . . , n is a projective
Riccati equation in P(n − 1, C). The equation corresponding to the second column
is a projective Riccati equation in P(n − 2, C(λi1 )), and so on.
Example 6.6. Let us compute the flag equation for the general differential
linear system of rank 3. Denote x = λ21 , y = λ31 , z = λ32 .

ẋ = a21 + (a22 − a11 )x + a23 y − a12 x2 − a13 xy
(6.4)
ẏ = a31 + a32 x + (a33 − a11 )y − a12 xy − a13 y 2
ż = a32 − a12 y + (a33 − a22 + a12 y − a13 y)z + (a13 y − a23 )z 2 .
6.7. Equations in the Special Orthogonal Group. Automorphic equa-
tions in special orthogonal group have been deeply studied since 19th century [44],
[10]. In particular Darboux related these equation with Riccati equation. He stated
that the integration of (6.5) is reduced to the integration of (6.11). Here we show
that the Flag equation of an automorphic equation in SO(C, 3) is precisely the Ric-
cati equation, and then the solutions of (6.5) are Liouvillian if and only if there are
algebraic solutions for (6.11)
The Lie algebra so(3, C) is the algebra of skew-symmetric matrices of gl(C, 3).
Then an automorphic system in SO(3, C) is written in the following form.
⎛ ⎞ ⎛ ⎞⎛ ⎞
ẋ0 a b x0
(6.5) ⎝ẋ1 ⎠ = ⎝−a c⎠ ⎝x1 ⎠ a, b, c ∈ K,
ẋ2 −b −c x2
where the void spaces represent the vanishing elements in the matrix.
6.8. On the Structure of the Special Orthogonal Group. The special

orthogonal group is the group of linear transformations preserving the quadratic
form x20 + x21 + x22 . Let us consider the non degenerated quadric in the projective
space S2 ⊂ P(3, C), defined by homogeneous equation {t20 + t21 + t22 − t23 = 0}. In
affine coordinates xi = tt3i , its affine part is a sphere of radius 1. Thus SO(3, C) is
a subgroup of algebraic automorphisms of the quadric; SO(3) ⊂ Aut(S2 ).
Each non degenerate quadric in the projective space over an algebraically closed
field is a hyperbolic ruled surface. It has two systems of generatrices, being each
system parameterized by a projective line. Denote P1 , P2 these projective lines.
p ∈ P1 , and q ∈ P2 are lines S2 , and they intersect in a unique point s(p, q) ∈ p ∩ q.
We have a decomposition of S2 which is a particular case of Segre isomorphism,
∼
P1 ×C P2 − → S2 ⊂ P(3, C)
⎧
⎪
⎪t0 = u0 v1 + u1 v0
⎨
t1 = u1 v1 − u0 v0
((u0 : u1 ), (v0 : v1 )) → (t0 : t1 : t2 : t3 )
⎪
⎪t 2 = i(u 1 v1 + u0 v0 )
⎩
t3 = u0 v1 − u1 v0
Let us consider any algebraic automorphism of S2 . τ : S2 → S2 . In particular,
it must carry a system of generatrices to a system of generatrices. Let us denote
P1 , P2 to the two system of generatrices of S2 . Hence, τ is induces by a pair of
projective transformations (τ1 , τ2 ), where
τ1 : P1 → P1 , τ2 : P2 → P2
or
τ1 : P1 → P2 , τ2 : P2 → P1 .
We conclude that the group of automorphism of S2 is isomorphic to the following
algebraic group,
Aut(S2 ) = P GL(1, C) ×C P GL(1, C) ×C Z/2Z.
Let us compute the image of the canonical monomorphism
SO(3, C) ⊂ Aut(S2 ). We take affine coordinates in the pair of projective lines,
x = uu01 , y = vv01 . This is the system of symmetric coordinates of the sphere intro-
duced by Darboux [10].
1 − xy 1 + xy x+y
(6.6) x0 = x1 = i x2 =
x−y x−y x−y
x0 + ix1 x2 − 1
(6.7) x= y= .
1 − x2 x1 − ix2
Let us write a general element of SO(3, C) in affine coordinates,
⎛ ⎞ ⎛ µ+µ−1 µ−1 −µ ⎞⎛ ⎞
1 1
λ−1 −λ ⎠ ⎜ µ−µ−1 ⎟⎝
2 2i
λ+λ−1 ν+ν −1 −1
Rλ,µ,ν = ⎝ 2 −1 2i−1 ⎝ µ+µ−1 ⎠ 2
ν −ν ⎠
2i−1
2i 2
λ−λ λ+λ ν−ν −1 ν+ν
2i 2 1 2i 2
where, in the complex case λ = eiα , µ = eiβ , ν = eiγ are the exponentials of the
Euler angles. Direct computation gives us,
!
x → (λµν+λν+µν−ν+λµ−λ+µ+1)x+λµν+λν+µν−ν−λµ+λ−µ−1
(λµν+λν−µν+ν+λµ−λ−µ−1)x+λµν+λν−µν+ν−λµ+λ+µ+1 = rλ,µ,ν (x)
Rλ,µ,ν
y → (λµν+λν+µν−ν+λµ−λ+µ+1)y+λµν+λν+µν−ν−λµ+λ−µ−1
(λµν+λν−µν+ν+λµ−λ−µ−1)y+λµν+λν−µν+ν−λµ+λ+µ+1 = rλ,µ,ν (y)
and then Rλ,µ,ν induces the same projective transformation rλ,µ,ν for x and y.
Hence,
SO(3) ⊆ P GL(1, C) ⊂ Aut(S2 ).
In particular, we have the following formulae for rotations around euclidean axis:
⎛ ⎞
1
⎝
−1 −1
λ −λ ⎠ (λ + 1)x + (λ − 1)
(6.8) λ+λ
: x →
2 −1
λ−λ λ+λ
2i−1 (λ − 1)x + (λ + 1)
2i 2
⎛ λ+λ−1 λ−1 −λ
⎞
2 2i
(6.9) ⎝ λ−λ−1 λ+λ−1 ⎠ : x → λx
2i 2
1
⎛ λ+λ−1 λ−1 −λ
⎞
−1 −1
⎠ : x → (λ + λ + 1/2)x − i(λ − λ )
2 2i
(6.10) ⎝ 1 −1 −1
i(λ − λ)x − (λ + λ + 1/2)
λ−λ−1 λ+λ−1
2i 2
An the following formulae for the induced Lie algebra morphism – the are computed
by derivation of previous formulae with λ = 1 + iε –. Here the Lie algebra pgl(1, C)
is identified with sl(2, C):
⎛ ⎞
1 i
⎝−1 ⎠ → 2 −i
0 2
⎛ ⎞
1 1

⎝ 0 ⎠ → 2
− 12
−1
⎛ ⎞
0
⎝ − 2i
1⎠ →
− 2i
−1
Reciprocally, a projective transformation
u11 x + u12 u11 y + u12
x → ; y → ,
u21 x + u22 u21 y + u22
induces a linear transformation in the affine coordinates x0 , x1 , x2 (see [10] p. 34).
SO(C, 3) is precisely the group of automorphisms of S2 that are linear in those
coordinates. We have proven the following proposition which is due to Darboux.
Proposition 6.7. The special orthogonal group SO(3, C) over an algebraically
closed field is isomorphic to the projective general group P GL(1, C). The isomor-
phism is given by formulae (6.8), (6.9), (6.10).
6.9. Flag Equation. The flag variety of SO(3, C) is a projective line. Any of
the Darboux symmetric coordinates,
x : S 2 → P1
gives us a realization of the action of SO(3) on P1 . By substituting the equation
(6.5) in the identities (6.6), (6.7) we deduce the Riccati differential equation satisfied
by this symmetric coordinate, which is the flag equation of equation (6.5):
−b − ic −b + ic 2
(6.11) ẋ = − iax + x .
2 2
In [10], Darboux reduces the integration of the equation (6.5) to finding two
different particular solutions of the Riccati equation (6.11). By application of our
generalization of Liouville’s theorem we obtain an stronger result.
Theorem 6.8 (Darboux). The Galois extension of the equation (6.5) is a Li-
ouvillian extension of K if and only if the Riccati equation (6.11) has an algebraic
solution.
Proof. It is a particular case of Theorem 6.2.
Appendix A. Stalk formula for affine morphisms

A.1. Stalk Formula for Ring Morphisms. Let us consider a ring mor-
phism ϕ : R → R , and a ⊂ R an ideal. We write ϕ(a) · R for the ideal of R
spanned by the image of a by ϕ.
Theorem A.1 (Stalk formula). Let us consider x ∈ Spec(R). The stalk

(ϕ∗ )−1 (x) ⊂ Spec(R ) is homeomorphic to the spectrum of
Rϕ(x)·R /ϕ(x) · Rϕ(x)·R = (R /ϕ(x) · R )ϕ(x)·R = R ⊗R κ(x).
Let us note that we do two different processes in the computation of the stalk.

First there is a process of localization: the spectrum of Rϕ(x)·R = R ⊗R Rx is

identified with the set of prime ideals y ⊂ R verifying ϕ(y) ⊆ x. Second there is a
process of restriction, the spectrum of R /ϕ(x) · R = R ⊗R R/x is identified with
the set of prime ideals y ⊂ R verifying ϕ(y) ⊇ x. These processes commute. When
we take both together we obtain R ⊗R κ(x). As expected, the canonical morphism
R → R ⊗R κ(x), a → a ⊗ 1 induces de immersion of the stalk into Spec(R ).
A.2. Stalk Formula for Change of Base Field.
Definition A.2. Let X be an k-scheme, and k → A a k-algebra. We write

X(A) for the set of k-scheme homomorphisms Spec(A) → X. The functor
X : A X(A) = Homk (Spec(A), X)

of the category of k-algebras in the category of sets, is called the functor of points
of X. An element x ∈ X(A) is called an A-point of X.
First, note that for each field extension k → K there is a map,

X(K) → X, x → x((0)), (0) ⊂ K
following this map, X(k) is identified with the set of points of X whose rational
field κ(x) is k. We call these points rational points of X.
For any field extension k ⊂ K, the map X(K) → X is surjective onto the
subset of points x ∈ X for whom that there exist a commutative diagram,
k BB /K
{=
BB {{
BB {
BB {{
{{
κ(x)
and moreover, X(K) is identified with the set of K-rational points of the K-scheme
XK :
X × Spec(K)
o7
ooooo
oo
ooo
Spec(K) /X
If X is of finite type, then X(k̄) → |X|cl ⊂ |X| is surjective onto the subset of
closed points of X.
Theorem A.3. There is a canonical one-to-one correspondence between the set
X(K) of K-points of X and the set of rational points of the extended scheme XK .
Proposition A.4 (Base change formula). Let X be a k-scheme, x ∈ X, and
k ⊂ A a k-algebra. The stalk π −1 (x) of x by π : XA → X, is isomorphic to
Spec(κ(x) ⊗k A).
Proof. First, assume that X = Spec(B) is affine. Then, by stalk formula, we
have
π −1 (x) = Spec(A ⊗k B ⊗B ⊗κ(x)) = Spec(A ⊗k κ(x)),
the homeomorphism is induced by the ring morphism
A ⊗k B → A ⊗ κ(x), a ⊗ f → a ⊗ f (x).
If X is not affine, then we cover it with affine subsets Ui . If π(y) = x, and
x ∈ Ui , then y ∈ Ui ×k Spec(A) and the previous argument is sufficient.
Acknowledgements. This research of both authors has been partially fi-
nanced by MCyT-FEDER Grant MTM2006-00478 of spanish goverment. The first
author is also supported by Civilizar, the research agency of Universidad Sergio
Arboleda. We also acknowledge prof. J.-P. Ramis and prof. E. Paul for their sup-
port during the visit of the first author to Laboratoire Emile Picard. We are also
in debt with J. Muñoz of Universidad de Salamanca for his continuous help and
support. We thank also P. Acosta, T. Lazaro and C. Pantazi who shared with us
the seminar of algebraic methods in differential equations in Barcelona. Finally, we
want to thank to the anonymous referee for his suggestions.
References
[1] M. Adler, P. van Moerbeke, P. Vanhaecke,“Algebraic Complete Integrable Systems,
Painlevé Architecture and Lie Algebras”, Springer Verlag, 2002.
[2] F. Benoist,D-algebraic geometry, Preprint 2008.
[3] D. Blázquez-Sanz, J. Morales-Ruiz, Local and Global Aspects of Lie Superposition Theo-
rem, preprint.
[4] D. Blázquez-Sanz, J. Morales-Ruiz, Lie’s Reduction Method and Differential Galois The-
ory in the Complex Analytic Context, Preprint 2009.
[5] A. Bialynicki-Birula, On Galois theory of fields with operators, Amer. J. Math. 84 (1962),
pp. 89–109.
[6] R. L. Bryant, An introduction to Lie Groups and Symplectic Geometry, Lectures at the
R.G.I. in Park City (Utah) 1991.
[7] A. Buium, Differential Function Fields and Moduli of Algebraic Varietes. Lecture Notes in
Mathematics, Springer Verlag, 1986.
[8] J. F. Cariñena, J. Grabowski, G. Marmo, “Lie-Scheffers systems: a geometric approach,”
Napoli Series on Physics and Astrophysics. Bibliopolis, Napoli, 2000.
[9] Carra’ Ferro, Kolchin schemes, J. Pure and Applied Algebra 63 (1990), pp. 13–27.
[10] G. Darboux, “Leçons sur la théorie générale des surfaces, I, II,” Editions Jacques Gabay,
Sceaux, 1993, (Reprint).
[11] A. Grothendieck, “Schèmas en groupes, 1962–1964 (Group schemes),” Lecture Notes in
Mathematics 151, 152 and 153, Springer Verlag 1970.
[12] Mark van Hoeij, Jacques-Arthur Weil, An algorithm for computing invariants of differ-
ential Galois groups, J. Pure App. Algebra 117 & 118 (1997), pp. 353–379.
[13] Mark van Hoeij, Jean-Franois Ragot, Felix Ulmer, Jacques-Arthur Weil, Liouvil-
lian solutions of linear differential equations of order three and higher. J. Symbolic Comput.
28 (1999), no. 4-5, pp. 589–609.
[14] J. E. Humphreys, “Linear Algebraic Groups”, Graduate Texts in Mathematics, Springer
Verlag 1975.
[15] I. Kaplansky, “An introduction to differential algebra”, Hermann, Paris 1957.
[16] W. F. Keigher, Differential Schemes and Premodels of Differential Fields, J. of Algebra 79
(1982), pp. 37–50.
[17] W. F. Keigher, On the structure presheaf of a differential ring, J. Pure and Applied Algebra
27 (1883), pp. 163–172.
[18] E. R. Kolchin, Galois Theory of Differential Fields, Amer. J. Math., Vol. 75, No. 4. (Oct.,
1953), pp. 753–824.
[19] E. R. Kolchin, Differential Algebra and Algebraic Groups, Academic Press, New York 1973.
[20] Jerald J. Kovacic, An algorithm for solving second order linear homogeneous differential
equations, J. Symbolic Comput. 2 (1986), pp. 3–43.
[21] Jerald J. Kovacic, Differential Schemes, Proceedings of the International Workshop, Rut-
gers University, Newark. World Scientific Publishing Co., River Edge, NL, 2002.
[22] Jerald J. Kovacic, The Differential Galois Theory of Strongly Normal Extensions, Trans-
actions of the AMS, Vol. 355, Number 11, pp. 4475–4522
[23] Jerald J. Kovacic, Geometric Characterization of Strongly Normal Extensions, Transac-
tions of the AMS, Vol. 358, Number 9, pp. 4135–4157
[24] S. Lie, Sur les équations différentielles ordinaries, qui possèddent des systemes fondamentaux
d’integrales, Compt. Rend. Acad. Sci. Paris, T CXVI (1893), pp. 11233-1235.
[25] J. Liouville, Mémoire sur l’integration de une classe de équations différentielles du second
ordre en quantités finies explicités, J. Math. Pures Appl. 4 (1839) pp. 423–456.
[26] B. Malgrange, “Le grupoı̈de de Galois d’un feuilletage,” Monographie 38 vol 2 de
L’ensegnaiment mathématique (2001).
[27] J.J. Morales-Ruiz, “Differential Galois Theory and Non-Integrability of hamiltonian sys-
tems,” Progress in Mathematics 179, Birkhäuser (1999).
[28] D. Mumford, Abelian Varieties, Tata Institute of fundamental Research, Bombay 1970.
[29] K. Nishioka, Differential Algebraic function fields depending rationally on arbitrary con-
stants, Nagoya Math. J. Vol. 113 (1989), pp. 173–179.
[30] K. Nishioka, General solutions depending rationally on arbitrary constants, Nagoya Math.
J. Vol. 113 (1989), pp. 1–6.
[31] K. Nishioka, Lie Extensions, Proc. Jap. Acad., 73, Ser. A (1997).
[32] J.F. Ritt,“Differential Algebra,” Dover, 1950.
[33] M. Rosenlicht, A remark on Quotient Spaces, An. da Acad. Brasileira de Ciencies, v. 35,
n. 4, 1963.
[34] C. Sancho de Salas, “Grupos Algebraicos y teorı́a de invariantes”, Sociedad Matemática
Mexicana, 2001.
[35] J. P. Serre, “CohomologieGaloisienne”, Lecture Notes in Mathematics 5, Springer Verlag,
5th Ed. Rep. 1997, (1st Ed 1964).
[36] Y. Sibuya, “Linear Differential Equations in the Complex Domain: Problems of Analytic
Continuation.” Transl. of Math. Monogr. 82, Am. Math. Soc. Providence, Rodhe Island 1990.
[37] M. F. Singer, F. Ulmer, Galois Groups of Second and Third Order Linear Differential
Equations, J. Symbolic Comput. (1993) 11, pp. 1–36.
[38] M. F. Singer, F. Ulmer, Liouvillian and Algebraic Solutions of Second and Third Order
Linear Differential Equations, J. Symbolic Comput, (1997) 11, pp. 37–73.
[39] F. Ulmer, J. A. Weil, Note on Kovacic’s algorithm J. Symbolic Comput. 22, pp. 179–200.
[40] H. Umemura, Birrational automorphism groups and differential equations, Proc. Fraco-
Japanese colloquium on differential equations, Strasbourg, 1985.
[41] M. Vanderput, M. Singer, “Galois theory of linear differential equations”, Grundlehren

der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], 328.
Springer-Verlag, Berlin, 2003.
[42] E. Vessiot, Sur l’intégration des quations diffrentielles linéaires. Annales Scientifiques de
l’Ecole Normale Supérieure Sr. 3, 9 (1892), pp. 197–280.
[43] E. Vessiot, Sur une classe d’equations différentielles. Annales scientifiques de l’E.N.S., 10
(1893), pp. 53–64.
[44] E. Vessiot, Sur une classe systèmes d’équations différentielles ordinaires. Compt. Rend.
Acad. Sci. Paris, T. CXVI (1893), pp. 1112–1114.
[45] E. Vessiot, Sur les systèmes d’équations différentielles du premier ordre qui ont des systèmes
fondamentaux d’intégrales. Annales de la faculté des sciences de Toulouse Sr. 1, 8 no. 3 (1894),
pp. H1–H33.
[46] E. Vessiot, Sur la théorie de Galois et ses diverses généralisations. Annales scientifiques de
l’cole Normale Suprieure Sr. 3, 21 (1904), pp. 9–85.
[47] E. Vessiot, Sur la réductibilité des systèmes automorphes dont le groupe d’automorphie est
un groupe continu fini simplement transitif. Annales scientifiques de l’cole Normale Suprieure
Sr. 3, 57 (1940), pp. 1–60.
(D. Blazquez-Sanz) IMA, Escuela de Matemáticas, Universidad Sergio Arboleda

E-mail address: david.blazquez-sanz@usa.edu.co
(J.J. Morales-Ruiz) Universidad Politécnica de Madrid

E-mail address: juan.morales-ruiz@upm.es
Volume 509, 2010
Recent Trends on Two Variable Orthogonal Polynomials
Lidia Fernández, Francisco Marcellán, Teresa E. Pérez,

and Miguel A. Piñar
Abstract. In this contribution we present a survey concerning orthogonal

polynomials in several variables. We emphasize two questions in order to do
a comparison with the one variable case.
First, according to different orderings in the monomial basis {xn y m }n,m∈N ,
we analyze the existence of recurrence relations for the corresponding sequences
of orthogonal polynomials.
Second, we study spectral properties of second order partial linear dif-
ferential operators with polynomial coefficients. Thus we introduce classical
orthogonal polynomials in a extended sense. Some characterizations in terms
of linear functionals are shown.
Finally, as an interesting application, some recent results for Sobolev inner
products associated with a vector of measures supported on the plane, are
presented.
Contents
1. Introduction 60
2. Algebraic properties of orthogonal polynomials in two variables 61
3. Orthogonal polynomials in two variables and eigenfunctions of second
order partial differential equations 67
4. Extended definition of classical orthogonal polynomials in two variables. 72
5. Semiclassical orthogonal polynomials in two variables 76
6. Sobolev orthogonal polynomials in several variables 78
7. Open problems 83
1991 Mathematics Subject Classification. 42C05; 33C50.

Key words and phrases. Orthogonal polynomials in two variables, classical orthogonal
polynomials.
The first, third and fourth authors are partially supported by Ministerio de Ciencia y Tec-
nologı́a (MCYT) of Spain and by the European Regional Development Fund (ERDF) through the
grant MTM 2008–06689–C02–02, and Junta de Andalucı́a, Grupo de Investigación FQM 0229.
The second author was partially supported by Ministerio de Educación y Ciencia of Spain,
grant MTM 2006–13000–C03–02, and Comunidad de Madrid–Universidad Carlos III de Madrid,
grant CCG07-UC3M/ESP-3339.
2010
c 0000
c Mathematical
American (copyright Society
holder)
1
59
60
2 L. FERNÁNDEZ, F. MARCELLÁN, T. E. PÉREZ, AND M. A. PIÑAR
8. Acknowledgements 84
References 84
1. Introduction
For a measure supported on a subset of the plane we introduce two variable
orthogonal polynomial sequences taking into account the Gram–Schmidt orthog-
onalization process for several choices in the ordering of the canonical basis of
monomials. From them, we deduce three–term recurrence relations with matrix
coefficients that such polynomial sequences satisfy. A connection with matrix or-
thogonal polynomials is established according to [DGIM06].
Following an historical approach (see [KS67], [KKL97], and [Su99]) based
on the extension of the Routh–Bochner characterization of classical orthogonal
polynomials in one variable (Hermite, Laguerre, Jacobi, and Bessel), a construc-
tive approach of some families of two variable orthogonal polynomials which are
eigenfunctions of second order partial linear differential operators with polynomial
coefficients is given. Then, using standard techniques for the symmetrization of
partial differential operators, we can deduce the weight function as well as the
corresponding domain of orthogonality .
In the more general framework of the orthogonality associated with moment
functionals on the linear space of polynomials in two variables with real coefficients,
classical orthogonal polynomials are defined in terms of a matrix analogue of the
Pearson differential equation that such a functional satisfies. They can also be char-
acterized as the polynomial solutions of a matrix second order partial differential
equation (see [FPP05a]).
The structure of the manuscript is as follows. In Section 2 we analyze some
algebraic properties of polynomials in two variables, orthogonal with respect to a
linear functional. We focus our attention in their construction according to two
different ways of ordering in the monomial basis. A three term recurrence relation
and a Christoffel-Darboux formula are deduced. Section 3 deals with the study
of polynomial eigenfunctions of a second order linear partial differential operator.
Under a condition of admissibility, nine families appear. This problem was studied
by H. L. Krall and I. M. Sheffer, as well as by P. K. Suetin. Such polynomials are
called ”classical” orthogonal polynomials because they represent the natural exten-
sion to the two variable case of the classical orthogonal polynomials in one variable
according to the Routh-Bochner approach. We point out some characterizations
of such polynomials as well as we determine the corresponding weight functions.
On the othe r hand, taking into account some examples of non admissible partial
differential operators, in Section 4 we extend this definition of classical orthogonal
polynomials taking into account a new eigenproblem in terms of vectors with poly-
nomial entries. The connection with matrix Pearson-type equations is stated as
well as some characterizations following the same ideas than in the univariate case
are given. In Section 5, the concept of semiclassical linear functional is introduced
and some examples are shown. Section 6 is focussed in an overview on Sobolev or-
thogonal polynomials in several variables, with the presentation of some examples
on the unit ball. Finally, some open problems are proposed.
RECENT TRENDS ON TWO VARIABLE ORTHOGONAL POLYNOMIALS 61
3
2. Algebraic properties of orthogonal polynomials in two variables

One of the first difficulties in the theory of multivariate orthogonal polynomials
is that it is necessary to choose an order in the basic monomials. There are two
different approaches concerning this question. It is possible to order the monomials
according to the lexicographical order
1, x, x2 , . . . xn , . . . , y, x y, x2 y, . . . , xn y, . . . . . .
or the reverse lexicographical order
1, y, y 2 , . . . y n , . . . , x, x y, x y 2 , . . . , xy n , . . . . . .
as Delgado, Geronimo, Iliev, and Marcellán did in [DGIM06]. Note that none of
these orderings respects the total degree. The advantage of this ordering is that
the moment matrix is a structured matrix, i.e., it is a block Hankel matrix where
the blocks are themselves Hankel matrices.
On the other hand, the monomials can be ordered using the graded lexico-
graphical order, that is, they are ordered according to the total degree and the
monomials with the same total degree are ordered using lexicographical order. This
ordering was introduced by Jackson [Ja36] and later on used by many authors (see
[KS67, Koor75, Su99]),
1, x, y, x2 , x y, y 2 , . . . , xn , xn−1 y, . . . , y n , . . .
The advantages of this ordering is that the orthogonality relations for polynomials
of degree n do not affect the orthogonality relations for polynomials of lower degree.
In this case it is usual to write the monomials using the vector notation introduced
by Kowalski [Ko82a, Ko82b] and later on developed by Xu [Xu93, DX01],
⎛ n ⎞
⎛ 2⎞ x
x ⎜xn−1 y ⎟
x ⎜ ⎟
X0 = (1); X1 = ; X2 = ⎝x y ⎠ ; · · · Xn = ⎜ . ⎟ ; · · ·
y
y2 ⎝ . ⎠
.
yn
This vector notation simplifies the formulation of the properties satisfied by the
polynomials. Next we will show the algebraic properties of orthogonal polynomials
in both cases.
2.1. Orthogonal polynomials in lexicographical and reverse lexico-

graphical order.
Moment functionals. In this section, we consider moment matrices associated
with the lexicographical ordering which is defined by
(k, l) <lex (k1 , l1 ) ⇔ k < k1 or (k = k1 and l < l1 )
and the reverse lexicographical ordering defined by
(k, l) <revlex (k1 , l1 ) ⇔ (l, k) <lex (l1 , k1 ).
Denote Π (x, y) as the span{xi y j , 0 ≤ i ≤ n, 0 ≤ j ≤ m}. Let Ln,m be a linear
n,m
functional defined on Π2n,2m (x, y) by

Ln,m (xi y j ) = hi,j .
Then, hi,j is said to be the (i, j) moment of Ln,m and Ln,m a moment functional.
If we built the (n + 1)(m + 1) × (n + 1)(m + 1) matrix Hn,m for Ln,m with respect
62
to the monomial basis according to the lexicographical ordering, then it has the
special form
⎛ ⎞
H0 H1 ··· Hn
⎜ H1 H2 · · · Hn+1 ⎟
⎜ ⎟
Hn,m = ⎜ . . .. ⎟,
⎝ .. .. . ⎠
Hn Hn+1 · · · H2n
where each Hi is a (m + 1) × (m + 1) matrix of the form
⎛ ⎞
hi,0 hi,1 ··· hi,m
⎜ hi,1 hi,2 · · · hi,m+1 ⎟
⎜ ⎟
Hi = ⎜ . .. .. ⎟ , i = 0, . . . , 2n.
⎝ .. . . ⎠
hi,m hi,m+1 ··· hi,2m
Thus Hn,m is a block Hankel matrix where each block is a Hankel matrix so it has
a double Hankel structure. If the reverse lexicographical ordering is used instead
of the lexicographical ordering another moment matrix H̃n,m is obtained where the
roles of n and m are interchanged.
We say that the moment functional Ln,m is positive definite if
Ln,m (p2 ) > 0
for all nonzero p ∈ Πn,m . Likewise, the moment functional Ln,m is nonnegative
definite if Ln,m (p2 ) ≥ 0 for all p ∈ Πn,m . Notice that Ln,m is positive definite or
nonnegative definite if and only if its moment matrix Hn,m is positive definite or
nonnegative definite, respectively.
Orthogonal polynomials and recurrence relations. In [DGIM06], the properties
of two variable orthogonal polynomials are studied. Let LN,M : Π2N,2M → R
be a positive definite linear functional. They define the orthonormal polynomials
pln,m (x, y), 0 ≤ n ≤ N, 0 ≤ m ≤ M , 0 ≤ l ≤ m, by the equations
LN,M (pln,m xi y j ) = 0, 0 ≤ i < n and 0 ≤ j ≤ m, or i = n and 0 ≤ j < l,

LN,M (pln,m pln,m ) = 1,
and
n,l i,j
pln,m (x, y) = kn,m,l + kn,m,l xi y j .
(i,j)<lex (n,l)
n,l
With the convention kn,m,l > 0, the above equations uniquely specify pln,m . Poly-
nomials orthonormal with respect to LN,M but using the reverse lexicographical
ordering will be denoted by p̃ln,m . They are uniquely determined by the above
relations with the roles of n and m interchanged.
Set
⎛ 0 ⎞ ⎛ ⎞ ⎛ 0 ⎞ ⎛ ⎞
pn,m 1 p̃n,m 1
⎜p1n,m ⎟ ⎜ y ⎟ ⎜p̃1n,m ⎟ ⎜ x ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
Pn,m = ⎜ . ⎟ = Kn,m ⎜ . ⎟ and P̃n,m = ⎜ . ⎟ = K̃n,m ⎜ . ⎟ ,
⎝ .. ⎠ ⎝ .. ⎠ ⎝ .. ⎠ ⎝ .. ⎠
pm
n,m xn y m p̃m
n,m xn y m
5
where Kn,m is the (m + 1) × [(n + 1)(m + 1)] matrix given by
⎛ 0,0 0,1 n,0 ⎞

kn,m,0 kn,m,0 ··· kn,m,0 0 ···
⎜ 0,0 .. ⎟
⎜ kn,m,1 0,1
kn,m,1 ··· n,0
kn,m,1 n,1
kn,m,1 . ⎟
Kn,m =⎜
⎜ .. ..
⎟,
⎟
⎝ .. .. .. .. ⎠
. . . . . .
0,1
0,0
kn,m,m kn,m,1 ··· ··· ··· n,m
kn,m,m
and K̃n,m is a (n + 1) × [(n + 1)(m + 1)] matrix given similarly with the roles of n
(n,m)
and m interchanged. Let Π(k) be the vector space of k dimensional vectors with
entries in Πn,m (x, y). An inner product for the vectors can be introduced by
X, Y = LN,M (XY t ).
The polynomials Pn,m can be obtained in an alternative way as follows. If µ is

a measure supported in a subset Ω of the plane, such that
LN,M (p) = p(x, y)dµ(x, y),

Ω
we associate with µ an (m + 1) × (m + 1) matrix valued measure Mm+1 (x) defined

by
dMm+1 (x) = (1, y, . . . , y m )T dµ(x, y)(1, y, . . . , y m ).

Ω
Let denote by {Pnm }n≥0 the sequence of (m+1)×(m+1) matrix valued polynomials
such that
Pnm (x)dMm+1 (x)[Pkm (x)]T = δn,k Im+1 ,

Ω
where Pnm (x) = Am n m

n,n x + lower degree terms, and An,n is a lower triangular matrix
with strictly positive diagonal entries.
Thus
Pn,m = Pnm (x)(1, y, . . . , y m )T
and from it we get
Theorem 2.1. ([DGIM06]) Given {Pn,m } and {P̃n,m }, 0 ≤ n ≤ N , 0 ≤ m ≤

M , the following recurrence formulas hold:
x Pn,m = An+1,m Pn+1,m + Bn,m Pn,m + Atn,m Pn−1,m ,

Γn,m Pn,m = Pn,m−1 − Kn,m P̃n−1,m ,
1
Jn,m Pn,m = y Pn,m−1 + Jn,m
2
P̃n−1,m + Jn,m
3
P̃n−1,m−1 ,
Pn,m = In,m P̃n,m + Γtn,m Pn,m−1
64
where
An,m = xPn−1,m , Pn,m ∈ M(m+1)×(m+1)
Bn,m = xPn,m , Pn,m ∈ M(m+1)×(m+1)
1
Jn,m = yPn,m−1 , Pn,m ∈ Mm×(m+1)
2
Jn,m = −yPn,m−1 , P̃n−1,m ∈ Mm×n
3
Jn,m = −yPn,m−1 , P̃n−1,m−1 ∈ Mm×n
Γn,m = Pn,m−1 , Pn,m ∈ Mm×(m+1)
Kn,m = Pn,m−1 , P̃n−1,m ∈ Mm×n
In,m = Pn,m , P̃n,m ∈ M(m+1)×(n+1)
and Mh×k denotes the linear space of real matrices of size h × k. Similar formulas
hold for P̃n,m (x, y).
Some relations between these matrices and an algorithm to compute them can
be found in [DGIM06], as well as a characterization of the families {pln,m (x, y)}
that can be represented as a tensor product of two families of one variable orthog-
onal polynomials.
With these definitions an analogue of the Christoffel–Darboux formula for or-

thogonal polynomials in one variable can be given.
Theorem 2.2. (Christoffel–Darboux formula)
Ptn,m (x1 , y1 )An+1,m Pn+1,m (x, y) − Ptn+1,m (x1 , y1 )Atn+1,m Pn,m (x, y)
x − x1
n
= Ptk,m (x1 , y1 )Pk,m (x, y)
k=0
m
= P̃tn,j (x1 , y1 )P̃n,j (x, y)
j=0
A similar result holds for the reverse lexicographical ordering.

2.2. Orthogonal polynomials in graded lexicographical order.
Moment functionals. Let Π denote the linear space of real polynomials in two
variables, and Πn the subspace of polynomials of total degree not greater than n.
We briefly give some general properties and tools about bivariate orthogonal
polynomials. For an exhaustive description of this and other related subjects see,
for instance, [KS67, Ko82a, Ko82b, Xu93, KKL97, KKL98, Su99, DX01].
Let {µh,k }h,k≥0 be a double indexed sequence of real numbers, and let us define
a functional u : Π → R by means of the moments µh,k = u, xh y k , h, k = 0, 1, 2, . . .,
and extended by linearity. Then, we will say that u is a moment functional and
µh,k is the (h,k) moment of u.
For a given moment functional u with moments {µh,k }h,k≥0 , the determinant
∆n is defined as the determinant of order (n + 1)(n + 2)/2 whose first row consists
of the elements
µ00 , µ10 , µ01 , µ20 , µ11 , µ02 , . . . , µn0 , µn−1,1 , . . . , µ0n
7
and whose subsequent rows are obtained by adding the integers (i, j) to the above
subscript pairs, where (i, j) runs successively through the values
(i, j) = (1, 0), (0, 1), (2, 0), (1, 1), (0, 2), . . . , (n, 0), (n − 1, 1), . . . , (0, n).
That is,
µ00 µ10 µ01
∆0 = µ00 , ∆1 = µ10 µ20 µ11 , ···
µ01 µ11 µ0,2
µ00 µ10 µ01 ··· µn0 ··· µ0n

µ10 µ20 µ11 ··· µn+1,0 ··· µ1n
∆n = .. .. .. .. .. .. .. .
. . . . . . .
µ0n µ1n µ0,n+1 ··· µnn ··· µ0,2n
The moment functional u is said to be quasi–definite if and only if ∆n = 0 for
n ≥ 0. We say that the moment functional u is positive definite if
u, p2 > 0
for all nonzero p ∈ Π. If u is positive definite, then the determinant ∆n > 0 .
Orthogonal polynomials and recurrence relations. The definition of orthogonal
polynomial in this case depends on the total degree of the polynomial. In fact, we
say that a polynomial p ∈ Πn is orthogonal with respect to u if
u, p q = 0, ∀q ∈ Π, deg q < deg p.
Definition 2.3. A polynomial system (PS) is a sequence of vectors {Pn }n≥0
of increasing size such that
Pn = (Pn,0 , Pn−1,1 , . . . , P0,n )t ,
where {Pn,0 , Pn−1,1 , . . . , P0,n } are polynomials of total degree n independent mod-
ulus Πn−1 .
Definition 2.4. Let u be a moment functional. We will say that a PS {Pn }n≥0
is a weak orthogonal polynomial system (WOPS) with respect to u if
u, Pn Ptm = 0, n = m,
u, Pn Ptn = Hn , n = 0, 1, 2, . . . ,
where Hn ∈ Mn+1 is a nonsingular matrix.
In the particular case where Hn is a diagonal matrix, we will say that the WOPS
{Pn }n≥0 is an orthogonal polynomial system (OPS). Moreover, if Hn = In+1 , we
call {Pn }n≥0 an orthonormal polynomial system.
In addition, a WOPS is called a monic WOPS if every polynomial contains
only one monic term of highest degree, that is,
P̃h,k (x, y) = xh y k + R(x, y), h + k = n,
where R(x, y) ∈ Πn−1 .
The following proposition was proved in [KS67].
Proposition 2.5. For a moment functional u the following statements are
equivalent
(i) u is quasi–definite.
66
(ii) There exists a unique monic WOPS relative to u.

(iii) There exists an OPS relative to u.
When the linear functional is positive definite, it is always possible to construct
an orthonormal polynomial system.
Example 2.6. (Tensor products of orthogonal polynomials in one variable)
The first example of OPS in two variables is obtained by multiplying two se-
quences of OP in one variable. In fact, let {Rh }h≥0 and {Sk }k≥0 be two families of
polynomials in one variable orthogonal with respect to the weight functions ω1 , and
ω2 , respectively. The family of polynomials in two variables defined by the tensor
product of these two families, that is,
Ph,k (x, y) = Rh (x)Sk (y), h, k ≥ 0,
is an OPS with respect to the weight function ω(x, y) = ω1 (x)ω2 (y).
Example 2.7. (A method for generating OP of two variables)
In 1975, Koornwinder [Koor75] gave a general method for generating bivariate
orthogonal polynomials from univariate ones. Let ω1 (x) be a weight function on
the interval (a, b), and ω2 (y) be a weight function on the interval (c, d). Let ρ(x)
be a positive function on (a, b) such that one of the following hypothesis holds
• ρ is a polynomial of degree ≤ 1,
• ρ2 is a polynomial of degree ≤ 2, c = −d, and w2 is an even function on
(−d, d).
For each integer k ≥ 0, let {pn,k (x)}n≥0 denote a sequence of orthogonal poly-
nomials with respect to the weight function ρ2k+1 (x)ω1 (x). Let {qn (y)}n≥0 be
a sequence of orthogonal polynomials with respect to ω2 (y). Then, polynomials
Pn−k,k of two variables can be defined by
y
Pn−k,k (x, y) = pn−k,k (x) ρk (x) qk , 0 ≤ k ≤ n.
ρ(x)
They are orthogonal with respect to the weight function
ω(x, y) = ω1 (x)ω2 (ρ−1 (x)y)
in the region {(x, y) : a < x < b, cρ(x) < y < dρ(x)}. With this construction
Koornwinder gives bases for orthogonal polynomials on the square, on the unit
disk, on the simplex, and on a parabolic domain. This result can be used to get
OPS in other bounded and unbounded domains.
Let u be a quasi–definite moment functional, {Pn }n≥0 an OPS with respect to
u and Hn = u, Pn Ptn . Then the following three term recurrence relations hold.
Theorem 2.8. [Ko82a, Xu93] For n ≥ 0, there exist unique matrices An,1 ,
An,2 , Bn,1 , Bn,2 , Cn,1 ,and Cn,2 , such that
(2.1) x Pn = An,1 Pn+1 + Bn,1 Pn + Cn,1 Pn−1 ,
(2.2) y Pn = An,2 Pn+1 + Bn,2 Pn + Cn,2 Pn−1 ,
where
An,1 Hn+1 = u, xPn Ptn+1 An,2 Hn+1 = u, yPn Ptn+1
Bn,1 Hn = u, xPn Ptn Bn,2 Hn = u, yPn Ptn
t t
An,1 Hn+1 = Hn Cn+1,1 An,2 Hn+1 = Hn Cn+1,2 .
9
Moreover,
(2.3) rank An,i = rank Cn+1,i = n + 1, i = 1, 2
t
An,1 Cn+1,1
(2.4) rank = rank t = n + 2.
An,2 Cn+1,2
In particular, if u is positive definite and {Pn }n≥0 is an orthonormal polynomial
system with respect to u, it is clear that Cn,i = Atn−1,i , for i = 1, 2.
The converse of this result is given in the next analogue of Favard’s theorem
proved in [Xu93] and, in a refined version, in [Xu94].
Theorem 2.9. Let {Pn }n≥0 , P0 = 1, be an arbitrary PS. Then the following
statements are equivalent.
(i) There exists a linear functional u which is quasi–definite on Π and such
that {Pn }n≥0 is an OPS with respect to u.
(ii) For n ≥ 0, i = 1, 2, there exist matrices An,i , Bn,i , and Cn,i such that
(1) the polynomials Pn satisfy the three term relations (2.1) and (2.2),
(2) the matrices in the relation satisfy the rank conditions (2.3) and
(2.4).
Let {Pn }n≥0 be an OPS with respect to u, we can define the kernel function
Kn
n
Kn ((x1 , y1 ), (x, y)) = Ptk (x1 , y1 ) Hk−1 Pk (x, y).
k=0
Notice that Kn satisfies the reproducing property. This definition does not depend
on the particular basis. The following theorem is an extension of the Christoffel–
Darboux formula for several variables.
Theorem 2.10. (Christoffel–Darboux formula)
n
Ptk (x1 , y1 ) Hk−1 Pk (x, y)
k=0
Ptn+1 (x1 , y1 )Atn,1 Hn−1 Pn (x, y) − Ptn (x1 , y1 )Hn−1 An,1 Pn+1 (x, y)
=
x1 − x
Ptn+1 (x1 , y1 )Atn,2 Hn−1 Pn (x, y) − Ptn (x1 , y1 )Hn−1 An,2 Pn+1 (x, y)
=
y1 − y
3. Orthogonal polynomials in two variables and eigenfunctions of

second order partial differential equations
In 1967, Krall and Sheffer [KS67] studied second order partial differential equa-
tions having orthogonal polynomials as eigenfunctions. There, they study a special
class of partial differential equations called admissible. The partial differential equa-
tion
(3.1) a∂xx p + 2b∂xy p + c∂yy p + d∂x p + e∂y p = λ p
is said to be admissible if there exists an infinite sequence {λn }n≥0 such that for
λ = λn , there are no non–zero polynomials solution of degree less than n and there
68
are precisely n + 1 linearly independent solutions of degree n. Then they proved

that equation (3.1) is admissible if and only if the coefficients are
a(x, y) = ax2 + d1 x + e1 y + f1
b(x, y) = 2a x y + d2 x + e2 y + f2
(3.2) c(x, y) = ay 2 + d3 x + e3 y + f3
d(x, y) = g x + h1
e(x, y) = g y + h2
λn = n[(n − 1)a + g]
and g + na = 0 for n = 0, 1, . . . (or equivalently λn = λm for n = m).
They defined classical orthogonal polynomials in two variables as the sequences
of orthogonal polynomials {Ph,k }h,k≥0 such that every polynomial Ph,k , with h+k =
n, satisfies the second order partial differential equation (3.1) with polynomial co-
efficients (3.2). This equation depends only on the total degree of the polynomial
solution and, therefore, all the polynomials of total degree n satisfy the same equa-
tion.
Moreover, they classified, up to a linear change of variables, all orthogonal
polynomial systems that arise as eigenfunctions of admissible equations. So, they
find nine equations
(1) pxx + pyy − xpx − ypy = −np,

(2) xpxx + ypyy + (1 + α − x)px + (1 + β − y)py = −np,
(3) pxx + ypyy − xpx + (1 + α − y)py = −np,
(4) xpxx + pyy + (1 + α − x)px − ypy = −np,
(5) (x2 − 1)pxx + 2xypxy + (y 2 − 1)pyy + gxpx + gypy = λn p,
(6) 3ypxx + 2pxy − xpx − ypy = −np,
(7) (x2 + y)pxx + 2xypxy + y 2 pyy + gxpx + g(y − 1)py = λn p,
(8) (x2 − x)pxx + 2xypxy + (y 2 − y)pyy + ((α + β + γ + 3)x − (1 + α))px
+((α + β + γ + 3)y − (1 + β))py = λn p,
(9) x2 pxx + 2xypxy + (y 2 − y)pyy + g(x − 1)px + g(y − α)py = λn p,
The first four equations correspond to tensor products of orthogonal poly-
nomials in one variable (Hermite–Hermite, Laguerre–Laguerre, Hermite–Laguerre,
Laguerre–Hermite). Polynomials satisfying the fifth equation are orthogonal poly-
nomials with respect to the weight function
ω(x, y) = (1 − x2 − y 2 )(g−3)/2 , g>1
on the closed unit circle. These polynomials were studied by Hermite [He08] and
Didon [Di68] for particular values of g and later on by Appell and Kampé de Fériet
[AK26] for general g. There is a polynomial solution of equation (8), orthogonal
with respect to the weight function
ω(x, y) = xα y β (1 − x − y)γ , α, β, γ > −1
on the simplex. These polynomials were first studied by Appell and Kampé de
Fériet [AK26]. The other three equations has been studied by several authors
[Li88, KKL97, KKL98, Su99, Lee00, KLL01, AFPP09]. In [KLL01], the
11
authors proved that all the Krall and Sheffer equations have an OPS as solution that
can be obtained by the Koornwinder construction given in Example 2.7, except for
the equation (6). Krall and Sheffer showed that this equation has an OPS solution
which cannot be positive definite. Anthony du Rapau found the monic PS of
solutions to (6) in a closed form, which is a WOPS but not an OPS, in [AFPP09]
a WOPS associated with a quasi–definite moment functional is constructed using
a Rodrigues–type formula.
Krall and Sheffer observed that if u is the moment functional associated with
a classical orthogonal polynomial sequence, then u satisfies the compatibility con-
ditions

a ux + b uy = (d − ax − by )u
(3.3)
b ux + c uy = (e − bx − cy )u
where ux and uy denote the distributional partial derivatives of the functional u,
defined by
ux , p = −u, ∂x p, uy , p = −u, ∂y p, ∀p ∈ Π,
and a, b, c, d, e are the polynomial coefficients in (3.1).
In 1988, Littlejohn ([Li88]), established a new approach to Krall and Sheffer
results. He studied symmetry factors for the differential operator L
(3.4) L[v] ≡ a vxx + 2 b vxy + c vyy + d vx + e vy .
where a, b, c, d, e are the polynomial coefficients in (3.1) satisfying (3.2).
Let L∗ be the formal Lagrange adjoint of L, defined by
(3.5) L∗ [v] ≡ (a v)xx + (2 b v)xy + (c v)yy − (d v)x − (e v)y .
As usual, L is symmetric if L[v] = L∗ [v]. A function ω(x, y) is called a symmetry
factor for L if ω(x, y) L is symmetric.
Littlejohn proved that ω(x, y) is a symmetry factor for L if and only if ω
simultaneously satisfies the (compatibility) equations
(a ω)x + (b ω)y = d ω,
(3.6) (b ω)x + (c ω)y = e ω.
Obviously, the nontrivial solutions of the above system of partial differential equa-
tions provides the symmetry factors for L. Assuming the coefficients of L suffi-
ciently differentiable and α ≡ ac − b2 = 0, it is possible to solve (3.6), under some
hypothesis. In [KKL98, Su99], a necessary and sufficient condition was given for
a differential operator L to admit a symmetry factor is
α βy − αy β = α γx − αx γ,
where
β = c (d − ax − by ) − b (e − bx − cy ),
γ = −b (d − ax − by ) + a (e − bx − cy ).
Moreover, in [KKL98], the authors showed that, in the Krall and Sheffer case, the
existence of an OPS solution of (3.1) implies the existence of a symmetry factor, and
then, the nine classical cases founded by Krall and Sheffer have symmetry factors.
These symmetry factors were calculated by P. K. Suetin ([Su99]) in all of the cases,
and were used to find some Rodrigues–type formulas ([KKL98], [Su99]).
70
Many properties for classical orthogonal polynomials in the Krall and Sheffer
sense were obtained in [KKL97, KKL98]. In particular, they gave a two–variable
version of Al–Salam and Chihara’s characterization of classical orthogonal poly-
nomials in one variable, the so–called structure relation. Moreover, they prove a
characterization in terms of a system of Pearson–type equations for the moment
functional. In fact, they obtain the following
Theorem 3.1 ([KKL98]). Let {Pn }n≥0 be a OPS relative to a quasi–definite
moment functional u and L an admissible differential operator. The following state-
ments are equivalent
(i) {Pn }n≥0 satisfy equation (3.1);
(ii) u satisfies the compatibility conditions (3.3)
(a u)x + (b u)y = d u,
(b u)x + (c u)y = e u,
(iii) there are (n + 1) × (k + 1) matrices Fkn and Gnk for k = n − 1, n, n + 1,
such that
a ∂x Pn + b ∂y Pn = n
Fn+1 Pn+1 + Fnn Pn + Fn−1
n
Pn−1 , n ≥ 1,
b ∂x Pn + c ∂y Pn = Gnn+1 Pn+1 + Gnn Pn + Gnn−1 Pn−1 , n ≥ 1.
However, there does not exist an analogue of the Hahn characterization of
classical orthogonal polynomials, since we can not deduce the orthogonality of the
partial derivatives of the Krall and Sheffer classical polynomials. In this context,
A. S. Lyskova (see [Ly91]), posed and solved the following problem: determine
a(x, y), b(x, y), and c(x, y) such that partial derivatives of any order of polynomial
solutions of the partial differential equation (3.1) satisfy a partial differential equa-
tion of the same type. The author showed that this property holds if and only if
ay = cx = 0, and called this class of partial differential equations the basic class.
However, A. S. Lyskova did not discuss the orthogonality of partial derivatives of
orthogonal polynomials satisfying the partial differential equation (3.1). This kind
of equation has been studied in [KKL97]. In [LLY04], the authors classify, up to
a real change of variables, all partial differential equations (3.1) which are in the
basic class, and show that partial derivatives of any order of orthogonal polynomial
solutions to the partial differential equations in the basic class are also orthogonal.
In [Su99], Suetin revisited the work of Krall and Sheffer, and he considered
fifteenth types of admissible partial differential equations (3.1) with polynomial
coefficients satisfying (3.2), deduced from the characteristic polynomial α = a c −
b2 . Some of the types described by Suetin are affine equivalent, however, these
types of admissible equations are essentially different in their properties: symmetry,
symmetry factors, orthogonality domain, weight function, Rodrigues formula.
Despite of all this extensive work, one has the feeling that the Krall and Sheffer
classification of classical orthogonal polynomials could be incomplete, since there
are non admissible partial differential equations having orthogonal polynomial solu-
tions. The most illustrative example is the OPS constructed by the tensor product
of Jacobi polynomials,
(α,β,α̂,β̂) (α,β) (α̂,β̂)
Ph,k (x, y) = Ph (x)Pk (y).
13
They are orthogonal on [−1, 1] × [−1, 1] with respect to the weight function
w(x, y) = (1 − x)α (1 + x)β (1 − y)α̂ (1 + y)β̂ , α, β, α̂, β̂ > −1.
Moreover, they satisfy the second order partial differential equation
(1 − x2 )pxx + (1 − y 2 )pyy + [β − α − (α + β + 2)x] px

+ β̂ − α̂ − (α̂ + β̂ + 2)y py = λh,k p,
where the coefficient of the term without derivatives depends on the partial de-
grees of the polynomial solution, and consequently, it is not an admissible partial
differential equation.
In [Koor75] three more examples are given, using Jacobi polynomials in one
(α,β)
variable Pn (x).
• The polynomials
(x−1/2 y),
(α,β) (α,β+k+1/2) (β,β)
Pn−k,k (x, y) = Pn−k (2x − 1) xk/2 Pk 0 ≤ k ≤ n,
are orthogonal with respect to the weight function (1−x)α (x−y 2 )β on the
region {(x, y) : y 2 < x < 1}. Moreover, they satisfy the partial differential
equation,
1
2x(1 − x) pxx + 2y(1 − x) pxy + (1 − x) pyy
2
+(−(2α + 2β + 5)x + 2β + 3)px − (α + 1)ypy = λn,k p.
• Another example is given using the weight function
ω(u, v) = (1 − u + v)α (1 + u + v)β (u2 − 4v)γ
for α, β, γ > −1, α + γ + 3/2 > 0, β + γ + 3/2 > 0 on the region
{(u, v) /|u| < v + 1, u2 − 4v > 0}.
The orthogonal polynomials obtained using the Gram–Schmidt procedure
satisfy the partial differential equation,
(u2 − 2u − 2) puu + 2(u v − u) puv + (−u2 + 2v 2 + 2v) pvv
+d(u, v)pu + e(u, v)pv = λn,k p,
where d(u, v) = 2(α − β) + u(3 + α + β + 2γ) and e(u, v) = 1 + u(α − β) +
2γ + v(5 + 2α + 2β + 2γ).
• In the same way as in the previous example, the system of orthogonal
polynomials with respect to the weight function
ω(x, y) = [−(x2 + y 2 + 9)2 + 8(x3 − 3xy 2 ) + 108]α
on the region bounded by the Steiner’s hypocycloid
−(x2 + y 2 + 9)2 + 8(x3 − 3xy 2 ) + 108 = 0,
satisfies the partial differential equation,
(−3x2 + y 2 + 6x + 9) pxx + 2(−4x y − 6y) pxy
+(x2 − 3y 2 − 6x + 9) pyy − 2(6α + 5)xpx − 2(6α + 5)ypy = λn,k p.
Obviously, all these equation are non admissible in the Krall and Sheffer sense and
therefore, the condition of admissibility could be too strong.
72
4. Extended definition of classical orthogonal polynomials in two

variables.
In [FPP05a, FPP05b], the authors extended the concept of classical orthog-
onal polynomials in two variables to a wider framework, which, of course, includes
the Krall and Sheffer definition, all the tensor products of classical orthogonal
polynomials in one variable, the above considered Koornwinder examples, and even
more.
The key to extend the concept of classical orthogonal polynomials in two vari-
ables is the vector representation for orthogonal polynomials. Let L be a second
order partial differential operator
L[p] = a∂xx p + 2b∂xy p + c∂yy p + d∂x p + e∂y p
where a, b, and c are polynomials of degree ≤ 2, and d, e are polynomials of degree
1 (without any particular shape) and let {Pn }n≥0 be a polynomial system. The
Krall and Sheffer equation (3.1) can be written in a matrix form as
L[Pn ] ≡ a ∂xx Pn + 2b ∂xy Pn + c ∂yy Pn + d ∂x Pn + e ∂y Pn = Λn Pn
where Λn = λn In . If we want to write the Koornwinder equations in a matrix form,
the formulation is the same but the matrix Λn is not a scalar matrix but a diagonal
matrix. This fact motivates the following definition that includes all the previous
examples.
Definition 4.1. Let u be a quasi–definite moment functional and {Pn }n≥0
the corresponding monic WOPS. Then u is classical (in an extended sense) if there
exist nonsingular matrices Λn ∈ M(n+1)×(n+1) such that
(4.1) L[Pn ] ≡ a ∂xx Pn + 2b ∂xy Pn + c ∂yy Pn + d ∂x Pn + e ∂y Pn = Λn Pn .
Remark 4.2. If Λn is a diagonal matrix, with entries λh,k , h+k = n, then every
orthogonal polynomial of total degree n, Ph,k , is an eigenfunction of the differential
operator L, that is
L[Ph,k ] = λh,k Ph,k , h, k ≥ 0, h + k = n.
The tensor products of classical orthogonal polynomials in one variable (Hermite,
Laguerre, Jacobi, and Bessel) and the Koornwinder examples correspond to this
situation. Moreover, if Λn = λn In+1 , is a scalar matrix, then every orthogonal
polynomial of total degree n satisfies the same partial differential equation. This is
the case studied by Krall and Sheffer ([KS67]) and Suetin ([Su99]).
The authors usually write equation (4.1) in terms of the gradient, ∇, and
divergence, div, operators defined as usual:

∂x p p
∇p = , div = ∂x p + ∂y q,
∂y p q
for all p, q ∈ Π. Using duality, we define the distributional gradient and the dis-
tributional divergence operators acting over moment functionals in the following
15
way

p p
∇u, = −u, div = −u, ∂x p + ∂y q, ∀p, q ∈ Π,
q q

∂x p
div(A u), p = −A u, ∇p = −u, At , ∀p ∈ Π,
∂y p
where A is a 2 × k polynomial matrix.
Defining

a b d
(4.2) Φ= , Ψ= ,
b c e
we can write
L[p] ≡ div (Φ∇p) + Ψ̃t ∇p,
where

d − ax − by
(4.3) Ψ̃ = Ψ − (div Φ) =t
.
e − bx − cy
Then, equation (4.1) can be expressed as
(4.4) L[Ptn ] ≡ div (Φ∇Ptn ) + Ψ̃t ∇Ptn = Ptn Λtn .
Remark 4.3. Notice that the definition of a classical family does not depend
on the particular choice of the monic WOPS. In fact, a matrix partial differential
equation equivalent to (4.1) is satisfied by every WOPS associated with a classical
moment functional u. Let {Pn }n≥0 be the monic WOPS associated with a classical
moment functional u, and let {Qn }n≥0 be another WOPS associated with u. For
n ≥ 0, let An be the nonsingular matrix corresponding to the change of basis
Qn = An Pn , n ≥ 0. Then,
L[Qn ] = Λ̃n Qn ,
where Λ̃n = A−1
n Λn An , that is, Λn and Λ̃n are similar matrices ([FPP05b]).
4.1. Characterizations. The extended definition of classical orthogonal poly-

nomials in two variables allowed (see [FPP05a, FPP05b]) to obtain the analogue
of the usual characterizations for classical orthogonal polynomials in one variable in
the two dimensional case: matrix Pearson–type equation for the moment functional,
structure relation, orthogonality of the gradients. These characterizations involve
matrix polynomial coefficients, the usual gradient operator, and the divergence op-
erator. Some of these properties appear, for the Krall and Sheffer particular case,
in [KS67, Li88, KKL97, KKL98, Su99] as previously mentioned.
Theorem 4.4. Let u be a quasi–definite moment functional, and let {Pn }n≥0
be the monic WOPS associated with u. The following statements are equivalent
(i) Matrix partial differential equation: u is a classical moment func-
tional (in the extended sense), that is, there exist polynomial matrices Φ,
Ψ of degree 2 and 1, respectively, such that for all n ≥ 0, there exist
nonsingular (n + 1) × (n + 1) matrices Λn with constant entries such that
L[Ptn ] ≡ div (Φ∇Ptn ) + Ψ̃t ∇Ptn = Ptn Λtn .
74
(ii) Matrix Pearson–type equation: there exist polynomial matrices Φ, Ψ

of degree 2 and 1, respectively, such that
div (Φ u) = Ψt u,
and detu, Φ = 0.
(iii) Structure relation: there exists a polynomial matrix Φ of degree 2, such
that {Pn }n≥0 satisfies
Φ ∇Ptn = (I2 ⊗ Ptn+1 )Fn+1
n
+ (I2 ⊗ Ptn )Fnn + (I2 ⊗ Ptn−1 )Fn−1
n
, n ≥ 1,
where n
Fm is 2(m + 1) × (n + 1) constant matrix for m = n + 1, n, n − 1,
and ⊗ denotes the usual Kronecker product of matrices ([B97]).
(iv) Orthogonality of the gradients: there exists a polynomial matrix Φ of
degree 2, such that {∇Ptn }n≥1 satisfy the orthogonality relations
u, (∇Ptm )t Φ∇Ptn = Kn δn,m , n, m ≥ 1,
where Kn are non singular symmetric matrices of order n + 1.
Recently, in [AFPP08a], the characterization for classical moment function-

als in one variable in terms of the Stieltjes function associated with the moment
functional has been extended to the bivariate case. In fact, the authors proved that
given a quasi–definite moment functional u, then u is classical if and only if the
Stieltjes function in two variables associated with u satisfies a non–homogeneous
version of the matrix Pearson–type equation for u.
4.2. The Rodrigues formula. Rodrigues formula for orthogonal polynomi-

als in two variables constitutes a quite old subject since it appears in the classical
monograph by Appell and Kampé de Fériet ([AK26]) for the ball and simplex
polynomials. Of course, the tensor product of two classical orthogonal polynomials
in one variable, Hermite, Laguerre, Jacobi, or Bessel polynomials, satisfies a Ro-
drigues formula, since they can be written as a product of the respective Rodrigues
formulas.
An analogue of the Rodrigues formula for other Krall and Sheffer classical
orthogonal polynomials in two variables has been obtained by Suetin ([Su99]). In
fact, for n a positive integer, he defines
1 n−i i n−i i
(4.5) Pn−i,i (x, y) = ∂ ∂y (p q ω),
ω x
where ω(x, y) is a weight function on a simply connected domain, and p(x, y),
q(x, y) are polynomials related to the polynomial coefficients in (3.1). Then, under
some additional hypothesis, the formula (4.5) defines an algebraic polynomial in
two variables orthogonal to all polynomials of degree not greater than n − 1. Kim,
Kwon, and Lee ([KKL98]) extend this formula to the non positive definite case.
However, in several cases, Rodrigues formula (4.5) provides polynomials of total
degree greater than n ([Su99]) and therefore it can not be used to construct a
basis of orthogonal polynomials.
In [AFPP09], the authors have obtained a matrix Rodrigues–type formula for
classical orthogonal polynomials in the extended sense. They gave a Rodrigues
distributional formula, which constitutes a generalization to the bivariate case of
the Cryer and Rasala results (see [Cr70] and [Ra81]), revisited in [MBP94], and
a matrix Rodrigues formula using a symmetry factor of the partial differential
17
equation (3.1). In fact, denoting by Φ{n} the second kind Kronecker power of
the matrix Φ (see [B97]), and, under some additional technical hypotheses, the
expression
1
(4.6) Qtn = div{n} (Φ{n} ω), n ≥ 0,
ω
provides a classical WOPS, where ω(x, y) is a symmetry factor of (3.1), and div{n}
is a n–th order differential operator.
It is important to remark that, for n ≥ 0, this Rodrigues formula generates
a (n + 1) vector of polynomials of total degree n orthogonal to all polynomials of
lower degree.
This formula holds for all of the classical cases described by Krall and Sheffer
paper ([KS67]), including classical orthogonal polynomials associated with a non
positive definite moment functional, as well as the tensor products of classical or-
thogonal polynomials in one variable. The matrix Rodrigues formula works even
in the cases where the Suetin formula fails, for instance, the formula allows the
construction of a WOPS in the case (6) of Krall and Sheffer. Moreover, the matrix
Rodrigues formula works also in some examples of classical orthogonal polynomials
(in an extended sense) not belonging to the Krall and Sheffer classification.
4.3. A non Krall and Sheffer case of classical orthogonal polynomi-

als. The extension of the definition of classical polynomials in two variables given
in [FPP05a, FPP05b] constitutes a more general framework, and it does not
reduce to the Krall and Sheffer cases and tensor products of classical orthogonal
polynomials in one variable. In [AFPP09], a new example of non–trivial classical
orthogonal polynomials in two variables is given.
Let us consider the moment functional u defined by means of its moments

(h + 4k)!
µh+4 k,h = u, xh+4 k y h = , (h, k = 0, 1, . . .),
k!
µh,k = u, x y = 0,
h k
otherwise.
As we can easily check, this moment functional is quasi–definite and it satisfies the
Pearson–type equation
div (Φ u) = Ψt u,
where
4 y2 1 −x
Φ= , Ψ= .
1 0 −y
Then, u is non positive definite and classical (in the extended sense), but it does
not fit into the Krall and Sheffer classification. Moreover, solving the Pearson–type
equation for a function ω(x, y), we obtain that a symmetry factor for the partial
differential operator is ω(x, y) = exp(y 4 − xy).
Then, applying the matrix Rodrigues formula (4.6) we get the WOPS
Qt0 = 1
Qt1 = (−x, −y)
2
Qt2 = x − 12y 2 , 2xy − 2, y 2
3
Qt3 = −x − 48y + 36xy 2 , −3x2 y + 36y 3 + 6x, −3xy 2 + 6y, −y 3
...
76
These weak orthogonal polynomials satisfy the matrix differential equation

L[Qtn ] = Qtn Λtn ,
where Λn are nonsingular upper triangular matrices:

−1 0
Λ0 = (0), Λ1 = ,
0 −1
⎛ ⎞
⎛ ⎞ −3 0 8 0
−2 0 8 ⎜ 0 −3 0 ⎟
24
Λ2 = ⎝ 0 −2 0 ⎠ , Λ3 = ⎜ ⎝ 0
⎟.
⎠
0 −3 0
0 0 −2
0 0 0 −3
5. Semiclassical orthogonal polynomials in two variables

Characterization (ii) in Theorem 4.4 for classical orthogonal polynomials is
the key to introduce the concept of semiclassical orthogonal polynomials in two
variables, as E. Hendriksen and H. van Rossum did in 1985 ([HR85]) in the one
variable case. In fact, in [AFPP07, AFPP08b] orthogonal polynomials in two
variables associated with semiclassical moment functionals are studied.
Definition 5.1. A quasi–definite moment functional u is said to be semiclas-
sical if it satisfies the matrix Pearson–type equation
(5.1) div (Φ u) = Ψt u,
where
a b d
Φ= , and Ψ= ,
b c e
are polynomial matrices with deg Φ = p ≥ 0, deg Ψ = q ≥ 1, and
(5.2) detu, Φ = 0.
We denote s = max{p − 2, q − 1} ≥ 0.
Remark 5.2. Notice that classical orthogonal polynomials in two variables
([KS67, Su99, FPP05a, FPP05b]) are semiclassical with s = 0. It is quite easy
to check that the tensor product of semiclassical orthogonal polynomials in one
variable yields semiclassical polynomials in two variables. Moreover, the Koorn-
winder construction (see Example 2.7) of bivariate orthogonal polynomials applied
to a pair of sequences of semiclassical orthogonal polynomials in one variable yields
also semiclassical orthogonal polynomials in two variables.
Semiclassical orthogonal polynomials in one variable can be characterized as the
only sequences of orthogonal polynomials satisfying one of the following equivalent
properties: the so–called structure relation, the quasi–orthogonality of the deriva-
tives, and a second order partial differential–difference relation (see for instance
[Ma87]). In [AFPP07], the bivariate analogue of the structure relation is proved,
and the other two characterizations are obtained in [AFPP08b].
Theorem 5.3. [AFPP07, AFPP08b] Let u be a quasi–definite moment func-
tional, and let {Pn }n≥0 be the monic WOPS associated with u. The following
statements are equivalent
(i) u is semiclassical.
19
(ii) Structure relation: {Pn }n≥0 satisfy

n+p−1
(5.3) Φ ∇Ptn = (I2 ⊗ Ptj )Fjn , for n ≥ s + 1,
j=n−s−1
where Fjn are real matrices of size 2(j + 1) × (n + 1).

(iii) Quasi–orthogonality relation for gradients: for n ≥ s + 1,
u, (∇Ptm )t Φ ∇Ptn = 0, 0 ≤ m < n − s.
(iv) Matrix differential–difference relation: there exist real matrices Λni
of size (i + 1) × (n + 1), such that
n+s
(5.4) L[Ptn ] ≡ div (Φ∇Ptn ) + Ψ̃t ∇Ptn = Pti Λni , n ≥ s + 1.
i=n−s
When n ≤ s, relation (5.4) reads

n+s
L[Ptn ] = Pti Λni ,
i=1
that is, Λn0 = 0, ∀n ≥ 0.
5.1. Examples. In [AFPP07, AFPP08b], some non trivial examples of
semiclassical orthogonal polynomials are given.
Example 5.4. In [AFPP07], using Koornwinder’s tools, an example of a semi-
classical weight function with unbounded support was given. For α, β > −1, the
polynomials
(x−1 y),
(α,β) (α+2k+1) (β,0)
Pn,k (x, y) = Ln−k (x) xk Pk n ≥ k ≥ 0,
are orthogonal with respect to the weight function
w(x, y) = xα e−x (1 − x−1 y)β ,
on the region {(x, y) : 0 < y < x}. Defining

x(x − y) 0
Φ= ,
0 x2 (x − y)
2
−x + x y + (α + 2)x + (β − α − 1)y
Ψ= ,
−(β + 1)x2
these polynomials are semiclassical. In this case, deg Φ = p = 3 and deg Ψ = q = 2,
so s = max{p − 2, q − 1} = 1, and the structure relation has, at most, p + s + 1 = 5
terms.
Example 5.5 (Appell–type polynomials ([AFPP08b])). Appell polynomials
([AK26]) are two variable orthogonal polynomials associated with the weight func-
tion
ω(x, y) = xα y β (1 − x − y)γ , α, β, γ > −1,
on the simplex
T 2 = {(x, y) ∈ R2 : x, y ≥ 0, 1 − x − y ≥ 0}.
The Appell moment functional u is defined as follows
u, f = f (x, y) ω(x, y) dx dy.

T2
78
Then, u is classical (i.e., it is semiclassical with s = 0), since it satisfies the matrix
Pearson–type equation (5.1), with

x(x − 1) xy (α + β + γ + 2)x − (α + 1)
Φ= , Ψ= .
xy y(y − 1) (α + β + γ + 2)y − (β + 1)
Appell–type polynomials can be defined as polynomials in two variables orthogonal
with respect to the moment functional
v = u + λ δ(x, y),
where λ ≥ 0 is a positive real number, and δ(x, y) is the usual Dirac distribution
at (0, 0). The action of v over polynomials is defined as follows,
v, f = f (x, y) ω(x, y) dx dy + λ f (0, 0).

T2
In [AFPP08b] v is proved to be semiclassical with s = 1, since v satisfies the

matrix Pearson–type equation (5.1)
(5.5) div (Φ̂ v) = Ψ̂t v,
where
Φ̂ = x Φ, Ψ̂ = (x(x − 1), x y)t + x Ψ.
6. Sobolev orthogonal polynomials in several variables

In the theory of orthogonal polynomials in one variable, the name Sobolev is
associated with polynomials that are orthogonal with respect to an inner product
involving both functions and their derivatives. This kind of polynomials has been
widely studied during the last twenty years and constitute the main subject of a vast
literature (see, for instance, [MF01, Ga04, MM06] and the references therein).
However, as far as we know, Sobolev orthogonal polynomials in several variables
have been studied in a few particular cases. In this moment, the only references in
the subject are [LL06, Xu06, Xu08, PX08].
6.1. Sobolev orthogonal polynomials in the unit ball. The first three
references ([Xu06], [Xu08] and [PX08]) are related to orthogonal polynomials on
the unit ball B d := {x : x ≤ 1} of the Euclidean space Rd .
Classical orthogonal polynomials on the unit ball are defined in terms of the
inner product
f, gµ = cµ f (x)g(x)(1 − x 2 )µ dx, µ > −1.

Bd
where cµ is the normalization constant of Wµ (x) = (1− x 2 )µ . The central symme-

try of the weight function Wµ plays an essential role in the construction of a basis
of mutually orthogonal polynomials, which can be expressed in terms of spherical
harmonics. Recall that spherical harmonics are homogeneous polynomials satisfy-
ing the Laplace equation ∆P = 0. If Hnd denotes the linear space of homogeneous
harmonic polynomials of degree n, it is well known that

n+d−1 n+d−3
dim Hn = σn =
d
− .
d−1 d−1
21
The spherical harmonics are the restriction of Y ∈ Hnd to the unit sphere S d−1 :=
{x : x = 1}, and they are orthogonal on S d−1 . From now on, we will use the
notation {Yνn : 1 ≤ ν ≤ σn } to denote an orthonormal basis for Hnd , that is,
1
(6.1) Yµn (x )Yνm (x )dω(x ) = δµ,ν δn,m , 1 ≤ µ, ν ≤ σn .
ωd S d−1
We will also denote by Πdn the space of polynomials of degree n in d variables

and by Πdn the space of homogeneous polynomials of degree n. It is well known
that dim Πdn = n+d−1
d−1 .
Let Vnd (Wµ ) denote the space of orthogonal polynomials of degree n. The
following theorem provides the announced basis
Theorem 6.1 (see [DX01]). A mutually orthogonal basis for Vnd (Wµ ) is given
by
(µ,n−2j+ d−2
2 )
(6.2) n
Pj,ν (Wµ ; x) = Pj (2 x 2
− 1)Yνn−2j (x), 0 ≤ j ≤ n/2,
(α,β)
where Pj denotes the Jacobi polynomial of degree j, which is orthogonal with re-
spect to (1−x)α (1+x)β on [−1, 1], and {Yνn−2j : 1 ≤ ν ≤ σn−2j } is an orthonormal
basis for Hn−2j
d
.
Notice that each element of the basis is the product of a spherical harmonic
and a radial part given by a Jacobi polynomial with parameters depending on its
degree.
In [Xu06] the author considers an inner product motivated by an application in

the numerical solution of the nonlinear Poisson equation −∆u = f (·, u) on the unit
disk with zero boundary conditions (see [AH06]). This inner product is defined by
(6.3) f, g∆ = αd ∆[(1 − x 2 )f (x)]∆[(1 − x 2 )g(x)]dx,

Bd
where ∆ is the usual Laplace operator and αd = 1/(4d2 vol(B d )) so that 1, 1∆ = 1.
Let Vnd (∆) denote the space of orthogonal polynomials with respect to ·, ·∆ .
Again, the symmetry of the inner product allows the construction of a basis of
mutually orthogonal polynomials, which can be expressed in terms of spherical
harmonics. So, mimicking the previous theorem, Y. Xu looked for a basis in the
form
Qnj,ν (x) = qj (2 x 2 − 1)Yνn−2j (x), 0 ≤ j ≤ n/2,
where qj is a polynomial of degree j in one variable and {Yνn−2j : 1 ≤ ν ≤ σn−2j }
is an orthonormal basis for Hn−2j
d
.
Now, the radial part of the Sobolev orthogonal polynomials is again orthogonal
with respect to an univariate Sobolev inner product, which depends on the degree
of the polynomials. In fact, the polynomial qj is orthogonal with respect to the
inner product
1
(f, g)βj := (Jβj f )(s)(Jβj g)(s)(1 + s)βj ds,
−1
where βj = n − 2j + (d − 2)/2, and
(Jβj qj )(s) = (1 − s2 )qj (s) + (βj − 1 − (βj + 3)s)qj (s) − (βj + 1)qj (s).
80
Using this property, an explicit representation for the Sobolev orthogonal poly-
nomials can be obtained.
Theorem 6.2 ([Xu06]). A mutually orthogonal basis for Vnd (∆) is given by
Qn0,ν (x) = Yνn (x),
(2,n−2j+ d−2
2 )
Qnj,ν (x) = (1 − x 2 )Pj−1 (2 x 2
− 1)Yνn−2j (x), 1 ≤ j ≤ n/2,
(α,β)
where Pj denotes the Jacobi polynomial of degree j and {Yνn−2j : 1 ≤ ν ≤ σn−2j }
is an orthonormal basis for Hn−2j
d
.
The explicit formula of the basis given in Theorem 6.2 leads to the following
interesting result, which relates Vnd (∆) to orthogonal polynomials with respect to
W2 (x) = (1 − x 2 )2
(6.4) Vnd (∆) = Hnd ⊕ (1 − x 2 )Vn−2
d
(W2 ).
Also, the explicit formula can be used to study further properties of the orthogonal
basis. In particular, it turns out the orthogonal expansion of a function f in the
basis can be computed without involving the derivatives of f .
In [Xu08], the author analyzes orthogonal polynomials for another kind of inner
products on the ball. At this occasion, he considers inner products involving the
usual gradient operator ∇. In particular he deals with two Sobolev inner products
defined by
λ 1
f, gI := ∇f (x) · ∇g(x)dx + f (x)g(x)dωd , λ > 0,
ωd B d ωd S d−1
and
λ
f, gII := ∇f (x) · ∇g(x)dx + f (0)g(0), λ > 0,
ωd B d
where the normalizing constants are chosen in such a way that 1, 1I = 1, 1II = 1.
Let Vnd (I) and Vnd (II) denote the spaces of orthogonal polynomials with respect to
·, ·I and ·, ·II , respectively. Using the same construction as above, a family of
explicit orthonormal bases is constructed for both spaces. The basis in [Xu06] and
those constructed in [Xu08] depend on Jacobi polynomials. It is interesting to note
that the basis for ·, ·I and the one for ·, ·∆ have the same structure. They differ
only in the parameters of the Jacobi polynomials.
The explicit representation for the Sobolev orthogonal polynomials with respect
to ·, ·I is given in the next theorem.
Theorem 6.3 ([Xu08]). A mutually orthogonal basis for Vnd (I) is given by
n
U0,ν (x) = Yνn (x),
(1,n−2j+ d−2
2 )
n
Uj,ν (x) = (1 − x 2 )Pj−1 (2 x 2
− 1)Yνn−2j (x), 1 ≤ j ≤ n/2.
Then, we get
(6.5) Vnd (I) = Hnd ⊕ (1 − x 2 )Vn−2
d
(W1 ).
For the inner product ·, ·II , the main result yields again an explicit family of
mutually orthogonal bases. The basis given for Vnd (II) turns out to be similar to
the basis for Vnd (I) given in Theorem 6.3. In fact, if n is odd, the two bases are
identical, whereas for n even, the two bases differ by just one element.
23
Theorem 6.4 ([Xu08]). A mutually orthogonal basis for Vnd (II) is given by
n−1
n
Vj,ν n
(x) = Uj,ν (x), 1 ≤ j ≤
2

1 (−1, d−2
2 )
n (d/2) n2
V n2n (x) = d−2
Pn (2 x 2
− 1) − (−1) 2
n+ 2 2 (n/2)!
where V nn (x) := V nn,ν (x) holds only when n is even.
2 2
An interesting consequence of those explicit formulas is that the Fourier ex-

pansion of a function f with respect to these orthogonal bases can be computed
without the use of the derivatives of f .
As it is well known, for µ > −1, orthogonal polynomials of degree n with

respect to the weight function Wµ (x) = (1 − x 2 )µ on the unit ball in Rd satisfy
a partial differential equation, which can be written in the following compact form

(6.6) ∆ − x, ∇2 − (2µ + d)x, ∇ P = −n(n + 2µ + d)P.
The singular case of the values µ = −1, −2, . . . is studied in [PX08]. Explicit
polynomial solutions are constructed and the equation for µ = −2, −3, . . . has
complete polynomial solutions if the dimension d is odd. An interesting result
obtained by the authors is that the orthogonal polynomials with respect to the
inner product
f, gI = λ ∇f (x) · ∇g(x)dx + f (x)g(x)dω, λ > 0,

Bd S d−1
which were studied in [Xu08], satisfy (6.6) for µ = −1.
6.2. The Lee & Littlejohn approach. In [LL06] the authors consider poly-
nomials in two variables which satisfy an admissible (as in (3.1)) second order partial
differential equation of the form
(6.7) L[v] ≡ avxx + 2bvxy + cvyy + dvx + evy = λv,
and are orthogonal with respect to a symmetric bilinear form defined by
(6.8) Φ(p, q) = σ, pq + τ, px qx ,
where a, b, c, d, e are polynomials in x and y satisfying conditions (3.2), λ is an
eigenvalue parameter, σ and τ are linear functionals acting on polynomials.
They find a condition for the partial differential equation (6.7) to have poly-
nomial solutions which are orthogonal with respect to a symmetric bilinear form
Φ(·, ·). The main result is provided by the following theorem.
Theorem 6.5. Let Φ(·, ·) be a symmetric bilinear form as in (6.8). The fol-
lowing statements (i) and (ii) are equivalent.
(i) The partial differential operator L[·] in (6.7) is symmetric with respect to
Φ in the linear space of polynomials, i. e.
Φ(L[p], q) = Φ(p, L[q]) ∀p, q ∈ Π,
82
(ii) σ and τ satisfy the relations

(aσ)x + (bσ)y − dσ = 0,
(6.9)
(bσ)x + (cσ)y − eσ = 0,

(aτ )x + (bτ )y − (d + ax )τ = 0,
(6.10)
(bτ )x + (cτ )y − (e + 2bx )τ = 0,
(6.11) cx τ = 0.
Furthermore, if {Pn }n≥0 is a Sobolev OPS with respect to Φ(·, ·), the statements (i)
and (ii) are equivalent to
(iii)) {Pn }n≥0 satisfies the partial differential equation (6.7).
Obviously, from the shape of the structure relations (6.9) and (6.10) we deduce
that the moment functionals σ and τ are closely connected. In fact, if both linear
functionals are quasi–definite, (6.9) and (6.10) implies that, under some additional
hypothesis there exists a polynomial f (x, y) of degree ≤ 2 such that τ = f (x, y)σ
and, if {Pn }n≥0 is a Sobolev OPS with respect to Φ(·, ·), then {Pn }n≥0 is a WOPS
with respect to σ and {∂x Pn }n≥0 contains a WOPS relative to τ .
However, the result does not reduce to the quasi–definite situation as we show
in the following example.
Example 6.6. Consider the differential equation
(6.12) xuxx + uyy + (1 + α − x)ux − yuy + nu = 0.
We know that (6.12) has a PS {Pn }n≥0 as solutions, with Pn = (pn,0 , . . . , p0,n )t ,
where every polynomial pn−k,k is the product of a Laguerre polynomial and an
Hermite polynomial. Then, by Theorem 6.5, σ and τ satisfy the relations

(xσ)x − (1 + α − x)σ = 0,
(6.13)
σy + yσ = 0,

(xτ )x − (2 + α − x)τ = 0,
(6.14)
τy + yτ = 0.
In the case α > −1, by solving (6.13) and (6.14), we have the distributional
representations for σ and τ

σ = xα e−x e− 2 y dxdy,
1 2
τ = xα+1 e−x e− 2 y dxdy.

1 2
In the case α = −1, we get the distributional representations for σ and τ

σ = δ(x) ⊗ e− 2 y dxdy,
1 2
τ = e−x e− 2 y dxdy.
1 2
Obviously in this case σ is not quasi–definite.

25
7. Open problems
Some open problems in the field of orthogonal polynomials in two variables will
be described in this section. We list only the most interesting open problems in our
opinion.
The first open problem is related with the classification of all of the classical
cases, in extended sense. We know the nine classes in the Krall and Sheffer clas-
sification [KS67], the additional classes of Suetin [Su99], the tensor products of
classical orthogonal polynomials in one variable, and the Koorwinder’s examples
in [Koor75]. Moreover, in [AFPP09] there is a new example not included in the
previous cases.
Problem 7.1. Classify the classical orthogonal polynomials in two variables
in the extended sense.
In [MBP94] classical orthogonal polynomials {Pn }n≥0 in one variable are char-
acterized by the fact that they satisfies
P (x) P (x) P (x)
Pn (x) = n+1 + an n + bn n−1 ,
n+1 n n−1
for n ≥ 2.
Problem 7.2. Is it true for classical orthogonal polynomials in two variables
in the extended sense?
Problem 7.3. Is there an analog of the Krall–Sheffer classification of second

order partial differential equations for several variables? For the discrete orthogonal
polynomials in several variables, the classification is given in [IX07], (see also there
for additional references on that topic.)
Problem 7.4. In [DGIM06] the one variable Bernstein–Szegő theory for or-
thogonal polynomials on the real line is extended to a class of two variable measures.
Is there an analog theory for the multivariate case?
The definition of semiclassical orthogonal polynomials in two variables as an

extension of the concept of classical orthogonal polynomials yields several problems.
Example 5.5 motivated the study of Krall–type polynomials in several vari-

ables. In [FPPX09], the authors analyzed algebraic properties for a bilinear form
obtained by adding a Dirac mass to a positive definite moment functional in the
linear space of polynomials in several variables.
Using the same notation as in example 5.5, let u be a positive definite moment
functional defined on Π. Let v be the moment functional defined as the following
perturbation of u
v, f = u, f + λ f (c), ∀f (x, y) ∈ Π,
where λ is a non zero real number and c ∈ R is a given point.
2
A first result in [FPPX09] gives a necessary and sufficient condition for the
existence of orthogonal polynomials with respect to the linear functional v. Explicit
84
formulas of orthogonal polynomials associated with v are derived from the orthog-
onal polynomials associated with the moment functional u. An explicit formula for
the reproducing kernel is also derived and used to establish certain inequalities for
classical orthogonal polynomials.
In one variable, it is well–known that Krall–type modifications of semiclassical
moment functionals yield semiclassical moment functionals, and the new Pearson–
type equation can be obtained from the original one. Using the above notations,
Problem 7.5. Let u be a semiclassical moment functional, and let v be the
Krall–type modification as above. Show that v is also semiclassical.
In the case of one variable, the study of this problem first aroses from the work
of A. M. Krall ([Kr81]) when he studied the sequences of orthogonal polynomials
that are eigenfunctions of a fourth order linear differential operator introduced in
1938–40 by H. L. Krall. He showed that the polynomials are orthogonal with respect
to a measure that is obtained from a continuous measure on an interval by adding
masses at the end points of the interval. In two variables, the tensor product of
Krall polynomials are eigenfunctions of a fourth order partial differential operator.
Problem 7.6. Study orthogonal polynomials in two variables that are eigen-
functions of fourth order partial differential operators.
Problem 7.7. In the one variable case, the factorization of the second order
linear differential equation with polynomial coefficients that a semiclassical sequence
of orthogonal polynomials satisfies, is related to the so called ladder operators.
What happens in the multivariate case?
8. Acknowledgements
The authors thank the careful revision of the manuscript by the referee. Their
comments and suggestions improved substantially the presentation.
References
[AFPP07] M. Álvarez de Morales, L. Fernández, T. E. Pérez, and M. A. Piñar, Semiclassical
orthogonal polynomials in two variables, J. Comput. Appl. Math. 207 (2007), 323–330.
[AFPP08a] M. Álvarez de Morales, L. Fernández, T. E. Pérez, and M. A. Piñar, A Stieltjes
function in two variables, in Approximation Theory XII: San Antonio 2007, Nashboro Press,
Vanderbilt (2008), 1–13.
[AFPP08b] M. Álvarez de Morales, L. Fernández, T. E. Pérez, and M. A. Piñar, A semiclassical
perspective on multivariate orthogonal polynomials, J. Comput. Appl. Math. 214 (2008),
447–456.
[AFPP09] M. Álvarez de Morales, L. Fernández, T. E. Pérez, and M. A. Piñar, A matrix Rodrigues
formula for classical orthogonal polynomials in two variables, J. Approx. Theory, 157 (2009),
32–52.
[AK26] P. Appell and J. Kampé de Fériet, Fonctions hypergéométriques et hypersphériques.
Polynômes d’Hermite, Gauthier-Villars, Paris, 1926.
[AH06] K. E. Atkinson and O. Hansen, Solving the nonlinear Poisson equation on the unit disk,
J. Integral Eq. and Appl. 17 (2006), 223–241.
[B97] R. Bellman, Introduction to matrix analysis, 2nd Edition, SIAM, Philadelphia, 1997.
[Cr70] C. W. Cryer, Rodrigues formulas and the classical orthogonal polynomials, Boll., Un. Mat.
Ital. (3) 25 (1970), 1–11.
[DGIM06] A. M. Delgado, J. S. Geronimo, P. Iliev, and F. Marcellán, Two Variable Orthogonal
Polynomials and Structured Matrices, SIAM J. Matrix Anal. and Appl., 28 (2006), 118–147.
27
[DGIX08] A. M. Delgado, J. S. Geronimo, P. Iliev, and Y. Xu, On a Two–Variable Class of

Bernstein-Szegő Measures, Constr. Approx., (2009), doi:10.1007/s00365-008-9022-2.
[Di68] F. Didon, Étude de certaines fonctions analogues aux fonctions Xn de Legendre, etc., Ann.
Sci. École Norm. Sup. 5 (1868), 229–310.
[DX01] C. F. Dunkl and Y. Xu, Orthogonal polynomials of several variables, Encyclopedia of
Mathematics and its Applications 81. Cambridge University Press, Cambridge, 2001.
[FPP05a] L. Fernández, T. E. Pérez, and M. A. Piñar, Classical orthogonal polynomials in two
variables: a matrix approach, Numer. Algorithms 39 (2005), 131–142.
[FPP05b] L. Fernández, T. E. Pérez, and M. A. Piñar, Weak classical orthogonal polynomials in
two variables, J. Comput. Appl. Math. 178 (2005), 191–203.
[FPPX09] L. Fernández, T. E. Pérez, M. A. Piñar, and Y. Xu, Krall–type Orthogonal Polynomials
in Several Variables, J. Comput. Appl. Math. (2009), doi:10.1016/j.cam.2009.02.067.
[Ga04] W. Gautschi, Orthogonal Polynomials: Computation and Approximation, Oxford Univ.
Press, Oxford, 2004.
[HR85] E. Hendriksen and H. van Rossum, Semi–classical orthogonal polynomials, in Polynômes
Orthogonaux et Applications, Bar–le–Duc 1984, C. Brezinski et al. Eds. Lecture Notes in
Math. 1171, Springer-Verlag, Berlin, 1985, 354–361.
[He08] C. Hermite, Oeuvres 2, Gauthier–Villars et Cie., Paris, 1908, 309–346.
[HJ91] R. A. Horn and C. R. Johnson, Topics in Matrix Analysis, Cambridge University Presss,
Cambridge, 1991.
[IX07] P. Iliev and Y. Xu, Discrete orthogonal polynomials and difference equations of several
variables, Adv. in Math. 212 (2007), 1–36.
[Ja36] D. Jackson, Formal properties of orthogonal polynomials in two variables, Duke Math. J.
2 (1936), 423–434.
[KKL97] Y. J. Kim, K. H. Kwon, and J. K. Lee, Orthogonal polynomials in two variables and
second–order partial differential equations, J. Comput. Appl. Math. 82 (1997), 239–260.
[KKL98] Y. J. Kim, K. H. Kwon, and J. K. Lee, Partial differential equations having orthogonal
polynomial solutions, J. Comput. Appl. Math. 99 (1998), 239–253.
[KLL01] K. H. Kwon, J. K. Lee, and L. L. Littlejohn, Orthogonal polynomial eigenfunctions of
second–order partial differential equations, Trans. Amer. Math. Soc. 353 (2001), 3629–3647.
[Koor75] T. Koornwinder, Two–variable analogues of the classical orthogonal polynomials, in
Theory and Application of Special Functions, R. Askey Editor, Academic Press (1975), 435–
495.
[Ko82a] M. A. Kowalski, The recursion formulas for orthogonal polynomials in n variables, SIAM
J. Math. Anal. 13 (1982), 309–315.
[Ko82b] M. A. Kowalski, Orthogonality and recursion formulas for polynomials in n variables,
SIAM J. Math. Anal. 13 (1982), 316–323.
[Kr81] A. M. Krall, Orthogonal polynomials satisfying fourth order differential equations, Proc.
Royal Soc. Edinburg Sect A 87 (1980/81), 271–288.
[KS67] H. L. Krall and I. M. Sheffer, Orthogonal polynomials in two variables, Ann. Mat. Pura
Appl. Serie 4 76 (1967), 325–376.
[Lee00] J. K. Lee, Bivariate version of the Hahn–Sonine theorem. Proc. Amer. Math. Soc. 128
(2000), 2381–2391.
[LLY04] J. K. Lee, L. L. Littlejohn, and B. H. Yoo, Orthogonal polynomials satisfying partial
differential equations belonging to the basic class, J. Korean Math. Soc 41 (2004), 1049–1070.
[LL06] J. K. Lee and L. L. Littlejohn, Sobolev orthogonal polynomials in two variables and second
order partial differential equations, J. Math. Anal. Appl. 322 (2006), 1001–1017.
[Li88] L. L. Littlejohn, Orthogonal polynomial solutions to ordinary and partial differential equa-
tions, in Proc. 2nd Intern. Symp. Orthogonal Polynomial and their Applications, M. Alfaro
et al. Editors, Segovia (Spain), 1986, Lect. Notes Math. 1329, Springer-Verlag, Berlin, 1988,
98–124.
[Ly91] A. S. Lyskova, Orthogonal polynomials in several variables, Soviet. Math. Dokl. 43 (1991),
264–268.
[MBP94] F. Marcellán, A. Branquinho, and J. Petronilho, Classical orthogonal polynomials: a
functional approach, Acta Appl. Math. 34 (1994), 283–303.
[MM06] F. Marcellán and J. J. Moreno-Balcazar, Asymptotics and zeros of Sobolev orthogonal
polynomials on unbounded supports, Acta Appl. Math. 94 (2006) 163–192.
86
[Ma87] P. Maroni, Prolégomènes à l’étude des polynômes orthogonaux semi–classiques, Ann. Mat.
Pura Appl. Ser. 4, 149 (1987), 165–184.
[MF01] A. Martı́nez–Finkelshtein, Analytic aspects of Sobolev orthogonality revisited, J. Comput.
Appl. Math. 127 (2001) 255–266.
[PX08] M. A. Piñar and Y. Xu, Orthogonal polynomials and partial differential equations on the
unit ball, Proc. Amer. Math. Soc. (2009) doi:10.1090/s0002-9939-09-09932-8.
[Ra81] R. Rasala, The Rodrigues formula and polynomial differential operators, J. Math. Anal.
Appl. 84 (1981), 443–482.
[Su99] P. K. Suetin, Orthogonal polynomials in two variables, Gordon and Breach, Amsterdam,
1999.
[Xu93] Y. Xu, On multivariate orthogonal polynomials, SIAM J. Math. Anal. 24 (1993), 783–794.
[Xu94] Y. Xu, Multivariable orthogonal polynomials and operator theory, Trans. Amer. Math.
Soc. 343 (1994), 193–202.
[Xu06] Y. Xu, A family of Sobolev orthogonal polynomials on the unit ball, J. Approx. Theory
138 (2006), 232-241.
[Xu08] Y. Xu, Sobolev orthogonal polynomials defined via gradient on the unit ball, J. Approx.
Theory 152 (2008), 52–65
(L. Fernández) Departamento de Matemática Aplicada, and Instituto Carlos I de

Fı́sica Teórica y Computacional, Universidad de Granada, Granada, Spain
E-mail address: lidiafr@ugr.es
(F. Marcellán) Departamento de Matemáticas, Universidad Carlos III de Madrid ,

Leganés, Madrid, Spain
E-mail address: pacomarc@ing.uc3m.es
(T. E. Pérez) Departamento de Matemática Aplicada, and Instituto Carlos I de

E-mail address: tperez@ugr.es
(M. A. Piñar) Departamento de Matemática Aplicada, and Instituto Carlos I de

E-mail address: mpinar@ugr.es
Volume 509, 2010
On the integrability of the Riccati equation
Cesar A. Gomez S.
Abstract. Using the Lie groups theory we show a method for integrating a
one-parameter family of Riccati equations. We obtain a new case of integration
of the general Riccati equation. Some applications are given.
Contents
1. Introduction 87
2. Symmetries to the one-parameter family of Riccati equations (1.2) 88
3. Applications 90
4. conclusions 92
References 93
1. Introduction
The simplest nonlinear ordinary differential equation is given by the general
Riccati equation (GRE)
dφ(ξ)
(1.1) = p(ξ)φ2 (ξ) + q(ξ)φ(ξ) + r(ξ),
dξ
where p(ξ), q(ξ), r(ξ) ∈ C[a, b], [a, b] ⊆ R and p(ξ) = 0. This equation is used in dif-
ferent fields of pure and applied mathematics, in theoretical physics, control theory
and relaxation problems [C],[CR],[S]. In special, solutions of particular cases of
(1.1) are used in a great variety of computational methods to obtain exact solutions
of nonlinear evolution equations (NEE) [G1],[GS1],[GS2],[GS3],[GS4],[GS5]. How-
ever, in spite of its apparent simplicity, the general solution to (1.1) cannot be ex-
pressed in an elementary form except in some particular cases [A],[D],[HS],[K],[M],[R],[S].
1991 Mathematics Subject Classification. Primary 35C05.

Key words and phrases. Lie group; Lie algebra; Riccati equation;
2010
c 0000
c Mathematical
holder)
1
87
88
2 CESAR A. GOMEZ S.
Starting from the one-parameter family of the Riccati equations

f (ξ)2 − 2f (ξ)f (ξ) + 4λ
(1.2) u (ξ) = E(ξ, u) = u2 (ξ) + ,
4f (ξ)2
where f (ξ) ∈ C 2 [a, b], f (ξ) = 0 and λ, an arbitrary parameter, we use the Lie
groups theory [B],[O],[O1] to obtain the general solution to the equation (1.2).
In accordance with the results, a new integrability condition to (1.1) is derived.
Finally, we give some applications.
2. Symmetries to the one-parameter family of Riccati equations (1.2)

Definition 2.1. The symmetries of (1.2) are given by the elements of a con-
nected Lie group with parameter a

ξ ∗ = ξ + ag(ξ, u) + o(a2 )
(2.1)
u∗ = u + ah(ξ, u) + o(a2 ),
which transform solutions into solutions.
Alternatively, the infinitesimal generators of the Lie algebra of (2.1), which are
the components of the vector field associated to (2.1) and given by
∂ = g(ξ, u)∂ξ + h(ξ, u)∂u ,
∂ ∂
where ∂ξ = ∂ξ , and ∂u = ∂u are called symmetries of (1.2). The symmetry variables
g(ξ, u), h(ξ, u) can be found solving the determinant equation [B],[O],[O1]
hξ (ξ, u) + (hu (ξ, u) − gξ (ξ, u))E(ξ, u)−
(2.2)
gu (ξ, u)E (ξ, u) − g(ξ, u)Eξ (ξ, u) − h(ξ, u)Eu (ξ, u) = 0.
2
In this direction, we have the following

Lemma 2.2. The one-parameter family of Riccati equations (1.2) admits the
vector field

f (ξ)
(2.3) ∂ = f (ξ)∂ξ − f (ξ)u + ∂u .
2
Proof. We seek solutions to (2.2) in the form
g(ξ, u) = f (ξ)
(2.4)
h(ξ, u) = k(ξ) + r(ξ)u.
Substitution of (2.4) into equation (2.2) and splitting with respect to u leads to the
following two equations
r(ξ) = −f (ξ)
(2.5) 1
k(ξ) = r (ξ).
2

f (ξ)
u) = −f (ξ)u
Therefore h(ξ, − 2 . It is easy to see that g(ξ, u) = f (ξ) and

h(ξ, u) = − f (ξ)u + f 2(ξ) satisfies (2.2).
ON THE INTEGRABILITY OF THE RICCATI EQUATION 89
3
Lemma 2.3. The solutions to the one-parameter family of Riccati equations

(1.2) are given by
1 1 1
u(ξ) = − ( f (ξ) + dξ ) if λ = 0,
f (ξ) 2 f (ξ)
√ √ dξ 1
λ tan( λ f (ξ) − c) − 2 f (ξ)
u(ξ) = if λ > 0,
f (ξ)
√ √ 1
− λ cot( λ fdξ (ξ) − c) − 2 f (ξ)
(2.6) u(ξ) = if λ > 0,
f (ξ)
√ √ 1
− −λ tanh( −λ fdξ (ξ) − c) − 2 f (ξ)
u(ξ) = if λ < 0,
f (ξ)
√ √ 1
− −λ coth( −λ fdξ (ξ) − c) − 2 f (ξ)
u(ξ) = if λ < 0,
f (ξ)
where c is an integration constant.
Proof. Solving the system

∂(t(ξ, u)) = 0
∂(w(ξ, u)) = 1,
where ∂ is given by (2.3), we obtain the so called canonical coordinates

t = t(ξ, u) = 12 f (ξ) + f (ξ)u
(2.7)
w = w(ξ, t) = fdξ (ξ)
.
In this new variables, (1.2) reduces to separable equation
dw 1
(2.8) = 2 .
dt t +λ
Using the solutions to (2.8) and taking into account (2.7), after simplifications we
obtain (2.6).
The following result can be considered as a new integrability condition to (1.1).
Theorem 2.4. If in (1.1), the coefficients are defined in any interval [a, b] ⊂ R
and p(ξ) ∈ C 2 [a, b], q(ξ) ∈ C 1 [a, b], r(ξ) ∈ C[a, b] are related as
1
·
4p2 (ξ)

− (p(ξ)q(ξ))2 + 4p(ξ)3 r(ξ)−
(2.9)
2p(ξ)p (ξ)q(ξ) − 3p2 (ξ) + 2p2 (ξ)q (ξ) + 2p(ξ)p (ξ) =
f 2 (ξ) − 2f (ξ)f (ξ) + 4λ

,
4f 2 (ξ)
with f (ξ) ∈ C 2 (a, b) , f (ξ) = 0, a properly chosen function, and λ, a properly
chosen constant, then the general solution to (1.1) can be obtain using quadratures.
90
4 CESAR A. GOMEZ S.
Proof. A calculation shown that with the substitution
p (ξ)
1 q(ξ) + p(ξ)
(2.10) φ(ξ) = u(ξ) − ,
p(ξ) 2
the equation (1.1) converts to (1.2). Using the Lemma 2.3 and the equation (2.10)
the solution to (1.1) is obtained.
3. Applications
For sake of simplicity, we consider only the following three examples
Example 3.1. The following is a first important illustrative example which is

used in some computational methods to obtain exact solutions to nonlinear par-
tial differential equations [G1],[GS1],[GS2],[GS3],[GS4],[GS5]. We consider the
Riccati equation
(3.1) φ (ξ) = γ(t)φ2 (ξ) + β(t)φ(ξ) + α(t),
where α(t), γ(t) = 0, β(t) are independent functions of ξ or arbitrary parameters.

Using (2.10), the equation (3.1) reduces to
4α(t)γ(t) − β 2 (t)
u (ξ) = u2 (ξ) + .
4
2
With λ = 4α(t)γ(t)−β
4
(t)
and f (ξ) = 1 we can use the Lemma 2.3. Therefore, by
(2.6) with c = 0 and using (2.10) we get the following set of solutions to (3.1):
(1) If α(t) = 0, γ(t) = 0 and β(t) = 0:

(3.2)
⎧ 1
⎪
⎪ (− 1ξ − β(t) β 2 = 4γ(t)α(t)
⎪
⎪ γ(t) 2 ),
⎪
⎪
⎪
⎪
⎪
⎪ √ √
⎪
⎪ 1
⎪
⎪
4α(t)γ(t)−β 2 (t) 4α(t)γ(t)−β 2 (t)
ξ] − β(t) 4α(t)γ(t) − β(t)2 > 0
⎪
⎪
( 2 tan[ 2 2 ),
⎪
⎪ γ(t)
⎪
⎪
⎪
⎪ √ √
⎨ 1
4α(t)γ(t)−β 2 (t) 4α(t)γ(t)−β 2 (t)
φ(ξ) = (− cot[ ξ] − β(t) ), 4α(t)γ(t) − β 2 (t) > 0
⎪
⎪ γ(t) 2 2 2
⎪
⎪
⎪
⎪
⎪
⎪ √ 2 √ 2
⎪
⎪ 1 β (t)−4α(t)γ(t) β (t)−4α(t)γ(t)
⎪
⎪ (− tanh[ ξ] − β(t)
2 ), 4α(t)γ(t) − β 2 (t) < 0
⎪
⎪ γ(t) 2 2
⎪
⎪
⎪
⎪
⎪
⎪ √ √
⎪
⎪ 1 (− β 2 (t)−4α(t)γ(t) coth[ β 2 (t)−4α(t)γ(t) ξ] − β ),
⎩ 2 2 2 4α(t)γ(t) − β 2 (t) < 0
γ(t)
5
(2) If α(t) = 0, γ(t) = 0:

⎧
⎪
⎪ β(t)
⎪
⎪ , β(t) = 0
⎪
⎪ −γ(t) + β(t)e−β(t)ξ
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ − , β(t) = 0
⎪
⎪ γ(t)ξ
⎪
⎪
⎪
⎪
⎪
⎪ √ 2 √ 2
⎪
⎪ 1 −β (t) −β (t)
⎪
⎪ ( tan[ ξ] − β(t) β 2 (t) < 0
⎪
⎨ γ(t) 2 2 2 ),
(3.3) φ(ξ) =
⎪
⎪ √ 2 √ 2
⎪
⎪ 1 −β (t) −β (t)
⎪
⎪ (− cot[ ξ] − β(t)
2 ), β 2 (t) < 0
⎪
⎪ γ(t) 2 2
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ (− β(t) β(t)
2 tanh[ 2 ξ] − 2 ),
β(t)
β 2 (t) > 0
⎪
⎪ γ(t)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎩ (− β(t) coth[ β(t) ξ] − β(t) ), β 2 (t) > 0
γ(t) 2 2 2
(3) If β(t) = 0, γ(t) = 0

⎧
⎪ 1
⎪
⎪ − , α(t) = 0
⎪
⎪ γ(t)ξ
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ ( α(t)γ(t) tan[ α(t)γ(t)ξ]), α(t)γ(t) > 0
⎪
⎪ γ(t)
⎪
⎪
⎪
⎪
⎪
⎨ 1
(3.4) φ(ξ) = (− α(t)γ(t) cot[ α(t)γ(t)ξ]), α(t)γ(t) > 0
⎪
⎪ γ(t)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ (− −α(t)γ(t) tanh[ −α(t)γ(t)ξ]), α(t)γ(t) < 0
⎪ γ(t)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎩ (− −α(t)γ(t) coth[ −α(t)γ(t)ξ]), α(t)γ(t) < 0
γ(t)
(4) If γ(t) = 0, β(t) = 0
−α(t) + β(t)eβ(t)ξ
(3.5) φ(ξ) = .
β(t)
Example 3.2. As a second illustrative example we construct an integrable fam-
2
ily of Riccati equations in the following form: setting λ = 0 and f (ξ) = eAξ +Bξ+C
in (1.2), we have the family
B 2 + 4A
(3.6) u (ξ) = u2 (ξ) − (Aξ 2 + ABξ + ),
4
which is integrable by quadratures according with the Lemma 2.3. It is easy to see
that the equations [K], [M]
(3.7) φ (ξ) = 1 + A(ξ − φ(ξ))φ(ξ),
92
6 CESAR A. GOMEZ S.
and
(3.8) φ (ξ) = φ2 (ξ) + ξφ(ξ) + ξ + 1,
after (2.10) are transformed in
A2 2 A
(3.9) u (ξ) = u2 (ξ) − ξ − ,
4 2
and
1 1
(3.10) u (ξ) = u2 (ξ) − ξ 2 + ξ − ,
4 2
respectively. Both equations belong to the family (3.6) and therefore are integrable
by quadratures using the Lemma 2.3. The equation (3.9) with A = A2 , B = 0, and
equation (3.10) with A = − 12 and B = 2.
Other illustrative example leading to many applications in mathematical physics
is given by the equation
(3.11) φ (ξ) + φ2 (ξ) = ξ 2 + 1.
Using (2.10) this last equation converts to
(3.12) u (ξ) = u2 (ξ) − ξ 2 − 1.
It is easy to see that (3.12) belong to the family (3.6) if we take A = 1 and B = 0.
In accordance with the first equation in (2.6) and using (2.10) we have the general
solution
1
φ(ξ) = ξ + −ξ2 −C .
e dξ
Note that φ(ξ) = ξ is a particular solution of (3.11). Using this, it is easy to
construct the general solution of the equation
φ (ξ) + φ2 (ξ) = ξ 2 + α,
where α = 2k + 1, and k = 1, 2, 3, . . . ,. We omit here.
Example 3.3. The general Riccati equation [K]
(3.13) φ(ξ) = Aξ m (φ(ξ)2 + 1),
satisfies the conditions of the Theorem 2.4 with f (ξ) = x−m and λ = A2 , therefore,
it is integrable by quadratures. The general solution can be obtained using (2.6)
and (2.10).
4. conclusions
In this work, using the Lie groups theory we have obtained the general solutions
to one-parameter family of Riccati equations. A new integrability conditions to the
equation (1.1) has been obtained. As we know, the theory of the Riccati equation is
related to the theory of the second-order ordinary differential equations, the results
obtained here can be adapted to this last case.
Acknowledgments: I would like to thank to referees for his helpful suggestions

and recommendations.
7
References
[A] J.L.Allen, F.M. Stein, On solutions of certain Riccati differential equations, Amer. Math.
Monthly 71, 1964, pp. 1113–1115 .
[B] G. Bluman, S. Kumey , Symmetries and Differential equations, Springer Verlag, New York
(1989).
[C] J.F Cariñena, G.Marmo, J.Nasare, The non-linear superposition principle and the Wei-
Norman method, Int.J.Mod.Phys ,1 1998, pp. 3601–3627.
[CR] J.F.Cariñena, A. Ramos, Lie systems and connections in fibre bundles: Applications in
Quantum Mechanics, Differential Geometry and its applications. Conf. Praga 2004. Charles
University, Prague (Zech Republic), 2005, pp. 437–452.
[D] H.T Davis, Introduction to non linear Differential and Integral Equations, Dover New York.
(1962).
[G1] C.A. Gomez, Special forms of the fifth-order KdV equation with new periodic and soliton
solutions, Appl. Math and Comp, 189 2007, pp. 1066–1077.
[GS1] C. Gomez, A. Salas, Exact solutions for a new integrable system (KdV6), J. Math. Science:
Advances and Appli, 1(2),2008, pp. 401–413.
[GS2] C.A. Gomez, A.H. Salas, The generalized tanh-coth method to special types of the fifth-order
KdV equation, Appl. Math and Comp, 203 (2008), pp. 873–880.
[GS3] A.H.Salas, C.A. Gomez, Computing exact solutions for some fifth KdV equations with
forcing term, Appl. Math and Comp, 204, 2008, pp.257–260.
[GS4] C.A. Gomez, A. Salas, Exact solutions for the generalized shallow water wave equation
by the general projective Riccati equations method, Boletin de Matematicas, XIII-1, 2006,
pp.50–56.
[GS5] C.A. Gomez, A. Salas, New exact solutions for the combined sinh-cosh-Gordon equation,
Lecturas Matematicas, special issue, 2006, pp. 87–93.
[HS] D.R.Haaheim, F.M. Stein, Methods of solution of the Riccati Differential Equation,
Matemathics Magazine, 42 No. 5, 1969, pp.233–240.
[K] E. Kamke, DifferentialGleichungen, Chelsea Publishing Company, New York, 1959.
[M] G. Murphy, Ordinary Differential Equations and their solutions, D. Van Nostrand Company
Inc, Princenton New Jersey 1960.
[O] P. J. Olver, Applications of Lie Group to Differential Equations. Springer-Verlag, 1980.
[O1] L.V Ovsianikov, Group Analysis of Differential Equations, Academic Press, New York, 1982.
[R] P.R.P Rao, V.H.Ukidave, Some separable forms of the Riccati equation, Amer. Math. Monthly
75 No. 10, 1968, pp.1113–1114.
[S] V.M.Strelchenya, A new case of integrability of the general Riccati equation and its application
to relaxation problems, J. Phys A. Math Gen 24, pp. 4965–4967.
Department of Mathematics, Universidad Nacional de Colombia, Bogotá, Colombia

E-mail address: cagomezsi@unal.edu.co
Volume 509, 2010
Two Discrete Systems of q-Orthogonal

Polynomials
∗
Mourad E.H. Ismail
October 29, 2009
Dedicated to the memory of Jairo Charris, a former student,

long time collaborator, and among the best of friends.
Abstract
We introduce two systems of orthogonal polynomials arising from birth
and death processes with exponentially decaying birth and death rates.
Their explicit formulas and orthogonality measures are found. Their con-
tinued J-fractions are also related to the Ramanujan continued fraction.
Running Title. Two Orthogonal Polynomials

Mathematics Subject Classification. Primary 33D45. Secondary 41A60.
Key words and phrases. Ramanujan continued fraction, birth and death
processes, Al-Salam–Ismail polynomials, new systems of orthogonal polynomials
continued J-fraction.
Filename: Ism:Jon.tex
1 Introduction
A birth and death process with birth rates {λn : n ≥ 0} and death rates
{µn : n ≥ 0} generates a sequence of orthogonal polynomials {Qn (x)} via
(1.1) Q0 (x) = 1, Q1 (x) = (λ0 + µ0 − x)/λ0 ,
and
(1.2) −xQn (x) = λn Qn+1 (x) + µn Qn−1 (x) − (λn + µn )Qn (x), n > 0.
It is assumed that
λn > 0, n ≥ 0, µn > 0, n > 0, µ0 ≥ 0.

∗ Research supported by a grant from King Saud University, Riyadh, Saudi Arabia
2010
c American Mathematical Society
1
95
96 MOURAD E. H. ISMAIL
When µ0 = 0 Ismail, Letessier and Valent [9] introduced another family of

polynomials {Q̃n (x)} generated by (1.2) and the initial conditions
(1.3) Q̃0 (x) = 1, Q̃1 (x) = (λ0 − x)/λ0 .
This means that we redefined µ0 to be zero. It is well-known that the measures of

orthogonality of {Qn (x)} and {Q̃n (x)} will be unique if λn and µn are bounded
sequences, [1].
In this paper we study the two families of polynomials corresponding to
bq 2n bq 2n−1
(1.4) λn = , µn = .
(1 + aq 2n )(1 + aq 2n+1 ) (1 + aq 2n )(1 + aq 2n−1 )
In section 2 we give explicit and asymptotic formulas for the two families of birth
and death process polynomials defined by the above birth and death rates. We
also show that they orthogonal with respect to a purely discrete and determine
its support and masses. This uses the techniques of [5]. Section 3 contains
related results for special values of the parameters. Our study uses the theory
of basic hypergeometric functions and many results from [2].
We shall follow the notation and terminology for basic hypergeometric series
as in [6] and [4]. In particular

n−1
(a; q)0 := 1, (a; q)n := (1 − aq k ), n > 0,
k=0
(1.5)

r
(a1 , a2 , . . . , ar ; q)n = (a; q)n .
k=1
We shall use Euler’s theorem

∞
(−z)n (n2 )
(1.6) q = (z; q)∞ .
n=0
(q; q)n
Al-Salam and Ismail [2] introduced the polynomials,
U0 (x; a, b) := 1, U1 (x; a, b) := x(1 + a),

(1.7)
x (1 + aq ) Un (x; a, b) = Un+1 (x; a, b) + bq n−1 Un−1 (x; a, b),
n
for q ∈ (0, 1), b > 0, a > −1. They showed that

n/2
(−a, q; q)n−k (−b)k xn−2k
(1.8) Un (x; a, b) = q k(k−1) ,
(−a, q; q)k (q; q)n−2k
k=0
and
(1.9) lim x−n Un (x; a, b) = (−a; q)∞ F (b/x2 ; a),

n→∞
TWO DISCRETE SYSTEMS OF Q-ORTHOGONAL POLYNOMIALS 97
see also §13.6 in [8]. Here F is defined by

∞
(−1)k xk q k(k−1)
(1.10) F (x; a) := .
(q, −a; q)k
k=0
Functions similar to F appear in the Rogers-Ramanujan identities [3], [10]. In

particular reference [10] establish asymptotics of the zeros of F (x; 0) as conjec-
tured by Ramanujan.
2 Two Systems of Orthogonal Polynomials

Let
U2n+1 (x; a, b)
(2.1) Rn (x2 ; a, b) := U2n (x; a, b), Sn (x2 ; a, b) :=
x(1 + a)
One can easily use (1.7) to see that
R0 (x; a, b) = 1, R1 (x; a, b) = (1 + a)(1 + aq) x − b,
(2.2) b
S0 (x; a, b) := 1, S1 (x; a, b) = (−aq; q)2 x − (1 + q)(1 + aq) .
1+a
Moreover the Rn ’s and Sn ’s satisfy the recurrence relations
Rn+1 (x; a, b) b2 q 4n−3 Rn−1 (x; a, b)
xRn (x; a, b) = +
(1 + aq 2n )(1 + aq 2n+1 ) (1 + aq 2n )(1 + aq 2n−1 )
(2.3)
bq 2n bq 2n−1
+ Rn (x; a, b) + ,
(1 + aq 2n )(1 + aq 2n+1 ) (1 + aq 2n )(1 + aq 2n−1 )
and
Sn+1 (x; a, b) b2 q 4n−1 Sn−1 (x; a, b)
xSn (x; a, b) = 2n+2 2n+1
+
(1 + aq )(1 + aq ) (1 + aq 2n )(1 + aq 2n+1 )
(2.4)
bq 2n+1 bq 2n
+ Sn (x; a, b) + ,
(1 + aq 2n+2 )(1 + aq 2n+1 ) (1 + aq 2n )(1 + aq 2n+1 )
respectively.
Let {Qn (x; a, b)} and {Q̃n (x; a, b)} be the polynomials corresponding to the
birth and death rates (1.4). We set
ψn (x; a, b) := (−1)n bn q n(n−1) Qn (x; a, b),

(2.5)
ψ̃n (x; a, b) := (−1)n bn q n(n−1) Q̃n (x; a, b)
It follows that {ψn (x; a, b)} and {ψ̃n (x; a, b)} solve the difference equation
yn+1 b2 q 4n−3 yn−1
xyn = +
(1 + aq 2n )(1 + aq 2n+1 ) (1 + aq 2n )(1 + aq 2n−1 )
(2.6)
bq 2n bq 2n−1
+ yn + .
(1 + aq 2n )(1 + aq 2n+1 ) (1 + aq 2n )(1 + aq 2n−1 )
The initial values of ψn (x; a, b) and ψ̃n (x; a, b) are
b(1 + q)(1 + a)
ψ0 (x; a, b) = 1, ψ1 (x; a, b) = x(1 + a)(1 + aq) −
(2.7) (q + a)
ψ̃0 (x; a, b) = 1, ψ̃1 (x; a, b) = x(1 + a)(1 + aq) − b
It readily follows from (2.6)–(2.7), (2.3)–(2.4), and (2.2) that
(2.8) ψn (x; aq, bq) = Sn (x; a, b), ψ̃n (x; a, b) = Rn (x; a, b).
Moreover the numerator polynomials [8] are given by
(2.9) ψn∗ (x; a, b) = ψ̃n∗ (x; a, b) = (1 + a)(1 + aq)ψn−1 (x; aq 2 , bq 2 ).
Formula (1.8) gives

n
(−a, q 2 ; q)2n−k (−b)k xn−k
(2.10) ψn (x; a, b) = q k(k−2) ,
(−a/q, q; q)k (q 2 ; q)2n−2k
k=0
n
(−a, q; q)2n−k (−b)k xn−k k(k−1)
(2.11) ψ̃n (x; a, b) = q .
(−a, q; q)k (q; q)2n−2k
k=0
Let µ and µ̃ be the measures with respect to which {ψn (x; a, b)} and {ψ̃n (x; a, b)}
are orthogonal. The Markov’s theorem [8], [13] gives

dµ(t; a, b) ψ ∗ (z; a, b) dµ̃(t; a, b) ψ̃ ∗ (z; a, b)
= lim n , = lim n ,
R z−t n→∞ ψn (z; a, b) R z−t n→∞ ψ̃ (z; a, b)
n
for z = 0.
Clearly (2.10)–(2.11) imply
(2.12) lim x−n ψn (x; a, b) = (−a; q)∞ F (b/xq; a/q),

n→∞
(2.13) lim x−n ψ̃n (x; a, b) = (−a; q)∞ F (b/x; a/q).

n→∞
There is no loss of generality is assuming b > 0.

Theorem 2.1. For a > −q and b > 0 the polynomials {ψn (x; a, b)} and
{ψ̃n (x; a, b)} are orthogonal with respect to probability measures µ and µ̃ whose
Stieltjes transforms are given by

dµ(t; a, b) F (bq/z; aq)
(2.14) = ,
z−t zF (b/zq; a/q)
R
dµ̃(t; a, b) F (bq/z; aq)
(2.15) = ,
R z−t zF (b/z; a/q)
respectively. Formulas (2.14) and (2.15) hold for z = 0. Hence µ and µ̃ are
purley discrete and have masses at the zeros of F (b/qz; a/q) and F (b/z; a/q),
respectively. The masses are the residues of the right-hand sides of (2.14) and
(2.15), at the respective mass points. Moreover we have the orthogonality rela-
tions

ψm (x; a, b)ψn (x; a, b) dµ(x) = ψ̃m (x; a, b)ψ̃n (x; a, b)dµ̃(x)
R R
(2.16)
1+a
= b2n q n(n−1) .
1 + aq 2n
Proof. The zeros of F (b/qz; a/q) and F (b/z; a/q) are real and simple for a > −q,
see [2]. We only need to show that the numerators and denominators in (2.14)
or (2.15) have no common zeros and that z = 0 does not support a mass. The
analysis is similar to that in [2] and will be omitted. Finally (2.16) follows from
the general results (5.2.13)-(5.2.14) and the last equation on page 137 of [8].
Note that Markov’s theorem [8], [13] [11], [12] shows that the continued
fraction expansion
F (bq/z; aq)
zF (b/zq; a/q)
(2.17)
1 λ 0 µ1 λn−1 µn
= ... ...,
x − λ0 − µ0 − x − λ1 − µ1 − x − λn − µn −
is valid in the complex plane except at the singularities of the right-hand side,
while
1 λ 0 µ1 λn−1 µn
... ...
x − λ0 − x − λ1 − µ1 − x − λn − µn −
(2.18)
F (bq/z; aq)
= ,
zF (b/z; a/q)
holds in C punctured at the singularities of the tight-hand side. In (2.17) and

(2.18) λn and µn are defined as in (1.4).
Hayman [7] studied the asymptotics of zeros of entire functions of the type
∞
2
(2.19) G(z) = q n an z n ,
n=0
where an = c[1 + O(q n )], as n → ∞. His results when applied to the function
F (z; a) show that xn (a; q) = cq −2n [1 + o(1)], as n → ∞, where
(2.20) 0 < x1 (a; q) < x2 (a; q) < · · · < xn (a; q) < · · · ,
are the zeros of F (z; a) and c is a constant. Therefore the point masses of the
measures in Theorem 2.1 converge to zero, as expected, and the location of the
nth point mass is O(q 2n ).
3 Some Special Cases

Let us first consider the polynomials {Un (x; q, b)}. As was observed in [2], the
application of Euler’s theorem (1.6) to F of (1.10) shows that
(1 + a)q
(3.1) F (z; q) = (z; q 2 )∞ , F (qz; q 2 ) = (z; q 2 )∞ − (z/q; q 2 )∞ .
z
One can also show that, [2],
(3.2) zF (zq; aq) = (1 + a)q [F (z; a) − F (z/q; a)] .
Therefore (3.1) yields
F (zq; q 3 ) = q 3 (1 + q)(1 + q 2 )z −2
(3.3)
× (1 + q − z/q 2 )(z/q; q 2 )∞ − (1 + q)(z/q 2 ; q 2 )∞ .
Therefore

dµ(t; q 2 , b) q 3 (1 + q)(1 + q 2 )z
=
(3.4) R z−t b2 (b/qz; q 2 )∞

× ((1 + q − bq −2 /z)(b/qz; q 2 )∞ − (1 + q)(bq −2 /z; q 2 )∞
Thus the measure µ(t; q 2 , b) is purely discrete and has masses at {xn }, and
(q 3 ; q 2 )∞ (q 3 ; q 2 )n 3n
(3.5) xn = bq 2n−1 , µ({xn }; q 2 , b) = q .
(q 6 ; q 2 )∞ (q 2 ; q 2 )n
The form of xn reaffirms the conclusions from Hayman’s theorem mentioned at
the end of §2. Similarly we have

dµ̃(t; q 2 , b) q 3 (1 + q)(1 + q 2 )z
=
(3.6) R z−t b2 (b/z; q 2 )∞

× ((1 + q − bq /z)(b/qz; q )∞ − (1 + q)(bq −2 /z; q 2 )∞
−2 2
the measure µ̃(t; q 2 , b) is purely discrete and has masses at {xn } with
xn = bq 2n ,
(3.7) (1 + q 2 )(q; q 2 )∞ (q 3 ; q 2 )n n
µ̃({xn }; q 2 , b) = 4 2 2 2
q − (1 + q)q 3n+2 .
(q ; q )∞ (q ; q )n
∞ ∞
The fact the n=0 µ({xn }; a, b) = n=0 µ̃({xn }; a, b) = 1 follows from the q-
binomial theorem ∞
(a; q)n n (az; q)n
z = ,
n=0
(q; q) n (z; q)n
[4], [6].
Acknowledgements This work resulted from a question raised by Bill

Jones at a speciall session on continued fractions at the American Mathematical
Society winter meeting in San Antonio, Texas in January 2006. I thanks Bill
for suggesting this problem, and Nancy Wyshinsky and Jimmy McLaughlin for
organizing such a stimulating special session.
References
[1] N. I. Akhiezer, The Classical Moment Problem and Some Related Ques-
tions in Analysis, English translation, Oliver and Boyed, Edinburgh, 1965
[2] W. Al-Salam and M. E. H. Ismail, Orthogonal polynomials associated with
the Rogers-Ramanujan continued fraction, Pacific J. Math. 105 (1983),
269–283.
[3] G. E. Andrews, q-series: Their development and application in analy-
sis, number theory, combinatorics, physics, and computer algebra, CBMS
Regional Conference Series, number 66, American Mathematical Society,
Providence, R.I. 1986.
[4] G. E. Andrews, R. A. Askey, and R. Roy, Special Functions, Cambridge
University Press, Cambridge, 1999.
[5] R. A. Askey and M. E. H. Ismail, Recurrence relations, continued fractions
and orthogonal polynomials, Memoirs Amer. Math. Soc. Number No 300
(1984).
[6] G. Gasper and M. Rahman, Basic Hypergeometric Series, second edition
Cambridge University Press, Cambridge, 2004.
[7] W. K. Hayman, On the zeros of a q-Bessel function, Contemporary Math-
ematics, volume 382, American Mathematical Society, Providence, 2005,
205–216.
[8] M. E. H. Ismail, Classical and Quantum Orthogonal Polynomials in one
Variable, Cambridge University Press, Cambridge, 2005.
[9] M. E. H. Ismail, J. Letessier and G. Valent, Linear birth and death models
and associated Laguerre and Meixner polynomials J. Approx. Theory 55
(1988), 337–348.
[10] M. E. H. Ismail and C. Zhang, Zeros of entire functions and a problem of
Ramanujan, Advances in Math. 209 (2007), 363–380.
[11] W. B. Jones and W. Thron, Continued Fractions: Analytic Theory and
Applications, Cambridge University Press, Cambridge, 1980.
[12] L. Lorentzen and H. Waadeland, Continued Fractions With Applications,
North-Holland, Amsterdam, 1992.
[13] G. Szegő, Orthogonal Polynomials, fourth edition, American Mathemati-
cal Society, Providence, 1975.
Department of Mathematics, University of Central Florida, Orlando, FL
32816
And
King Saud University, Riyadh, Saudi Arabia
email: ismailmath.ucf.edu
Volume 509, 2010
Like-Hyperbolic Bloch-Bergman classes
J. L
awrynowicz, L. F. Reséndis O., and L. M. Tovar S.
Abstract. In this paper we introduce the like-hyperbolic Bloch-Bergman

classes of bounded analytic functions in the open unit disk. We obtain for
them, several integral and series characterizations. Likewise we present some
metric properties and their relationships with some other well known classes.
Contents
1. Introduction. 103
2. Properties of B α∗
A 104
3. Characterizations for α-Bloch-Bergman classes 106
4. Carleson and series characterization of B α∗ α∗
A and B A,0 110
α∗
5. Properties of B A 112
References 1 16
1. Introduction.
Let 0 < r. Define Dr (a) := {z ∈ C : |z − a| < r} and Dr = Dr (0) . We denote
by D = D1 the open unit disk in the complex plane C and T its boundary. Let
φa : C → C be the Möbius transformation,
a−z
φa (z) = , |a| < 1,
1 − az
with pole at z = 1/a that verifies φ−1
a = φa . We observe that
(1 − |a|2 )(1 − |z|2 )

(1.1) 1 − |φa (z)|2 = = |φa (z)|(1 − |z|2 ) .
|1 − az|2
1991 Mathematics Subject Classification. Primary 30C45.

Key words and phrases. bounded analytic functions, Bloch-Bergman classes.
Partially supported by CONACYT and UAM 2230302.
Partially supported by CONACYT, Becario de COFAA-IPN.
1
103
101
104
102
2 J. L
AWRYNOWICZ, L. F. RESÉNDIS O., AND L. M. TOVAR S.
For z, a ∈ D, we denote Green’s function of D, with logarithmic singularity at a,

by
|1 − az| 1
(1.2) g(z, a) = ln = ln .
|z − a| |φa (z)|
Let B(D) be the set of all analytic functions f : D → D. In 2005 Xiaonan Li [Li]
introduced the hyperbolic Bloch class B, consisting of all functions f ∈ B(D) such
that
f B ∗ = sup(1 − |z|2 )f ∗ (z)2 ,
a∈D
where
|f (z)|
(1.3) f ∗ (z) =
1 − |f (z)|2
is the hyperbolic derivative [Ya]. Let 0 ≤ α < ∞ . We say that f ∈ B(D) belongs
to the like-hyperbolic Bloch Bergman class B α∗
A if
|f (z)|
f B α∗ = sup(1 − |z|2 )α <∞
A
z∈D 1 − |f (z)|2
and to the little like-hyperbolic Bloch Bergman class B α∗
A,0 if
|f (z)|
lim (1 − |z|2 )α =0.
|z|→1−1 1 − |f (z)|2
The aim of this paper is to obtain explicitely properties of the weighted like-
hyperbolic Bloch-Bergman classes. The main references are due to R. Aulaskari,
L. Reséndis, L. Tovar [AuReTo], X. Li [Li], and Ruhan Zhao [Zha], [Zha1].
2. Properties of Bα∗
A
In this part we clarify some elementary aspects of our function classes. It is

β∗
immediate from the definition that if 0 < α < β < ∞, then B α∗ α∗
A ⊂ B A , B A,0 ⊂
β∗
B A,0 . The following two results are fundamental for the theory developed in this
paper.
Lemma 2.1 (Yamasita, [Ya]). Let f ∈ B(D) . Then the function g : D → [0, ∞)
defined by
g(z) = ln[− ln(1 − |f (z)|2 )]
is subharmonic.
We need the following result.
Corollary 2.2. Let 0 < p < ∞ and f ∈ B(D). Then the function g : D →
[0, ∞) defined by
|f (z)|p
g(z) =
(1 − |f (z)|2 )2
is subharmonic.
Proof. It is well known that if f (z) is an analytic function then for 0 < p < ∞,
the function p ln |f (z)| is subharmonic. By Lemma 2.1, if f ∈ B(D) then z →
ln[− ln(1 − |f (z)|2 )] is a subharmonic function . Therefore since ex is a convex
LIKE-HYPERBOLIC BLOCH-BERGMAN CLASSES 103
105
3
function then z → −2 ln(1 − |f (z)|2 ) is a subharmonic function too. As the sum of

two subharmonic functions is subharmonic, then
|f (z)|p
p ln |f (z)| − 2 ln(1 − |f (z)|2 ) = ln
(1 − |f (z)|2 )2
is subharmonic and we apply again ex . This concludes the proof.
By the definition of hyperbolic Bergman classes it is clear the usefulness of the
following result (See Exercise 1, pag. 128 of [Co])
Theorem 2.3. Let f ∈ B(D). Then if f ∈ B(D) then
|f (0)| − |z| |f (0)| + |z|
≤ |f (z)| ≤ for all z ∈ D.
1 − |f (0)||z| 1 + |f (0)||z|
In particular
1 1 + |f (0)||z|
(2.1) ≤ for all z ∈ D.
1 − |f (z)| (1 − |f (0)|)(1 − |z|)
Outline of the Proof: Consider g : D → D defined by
f (z) − f (0)
g(z) =
1 − f (0)f (z)
and apply Schwarz’s Lemma.
∗
A := B A and if
Corollary 2.4. Let 1 ≤ α < ∞. If α = 1, then B(D) = B 1∗
α∗
1 < α < ∞ then B(D) = B A,0
Since f ∈ B(D) implies that f is bounded we have the following result:
Proposition 2.5. Suposse that f ∈ B(D) and 0 ≤ α < ∞ . Then f ∈ B α∗
A or
f ∈ B α∗
A,0 if and only if
(1 − |z|2 )α (1 − |z|2 )α
(2.2) sup <∞ or lim − =0
z∈D 1 − |f (z)|2 |z|→1 1 − |f (z)|2
respectively.
Proof. Let f ∈ B(D) . Since |f (z)| < 1 then (2.2) implies f ∈ B α∗ α∗
A or f ∈ B A,0
α∗ α∗
respectively. Reciprocally, for f ∈ B A or f ∈ B A,0 , define
1
Ω = { z ∈ D : |f (z)| ≥ },
2
therefore for z ∈ Ω we have
1 (1 − |z|2 )α |f (z)|
≤ (1 − |z|2 )α .
2 1 − |f (z)|2 1 − |f (z)|2
For z ∈ D \ Ω we obtain
(1 − |z|2 )α 4
≤ (1 − |z|2 )α .
1 − |f (z)| 2 3
From these estimations we get that (2.2) is satisfied respectively.
We need the following elementary estimations
106
104
4 J. L
Lemma 2.6. Let q ∈ R and a ∈ D. Then for all z ∈ D

1
(2.3) (1 − |z|2 )q ≤ (1 − |φa (z)|2 )q ≤ ρ(a, q)(1 − |z|2 )q
ρ(a, q)
where
|q|
1 + |a|
ρ(a, q) = .
1 − |a|
Corollary 2.7. Let q ∈ R and a ∈ D. Then for all z ∈ D
1
(2.4) ≤ |φa (z)|q ≤ ρ(a, q) .
ρ(a, q)
From these results we obtain that both classes B α∗ α∗
A and B A,0 are invariant under
conjugation by the Möbius group GM of automorphism of the unit disk.
Theorem 2.8. Let 0 < α < ∞. Then
GM ◦ B α∗ −1 α∗
A ◦ GM = B A , GM ◦ B α∗ −1 α∗
A,0 ◦ GM = B A,0 .
Proof. Let a ∈ D and f ∈ B α∗

A . Then by (1.1)
(1 − |z|2 )α (1 − |φa (z)|2 )α (1 − |z|2 )α

=
1 − |f ◦ φa (z)|2 1 − |f ◦ φa (z)|2 (1 − |φa (z)|2 )α .
Since |φa (z)| → 1− when |z| → 1− , we have by (2.4)
1 (1 − |z|2 )α
f B α∗ ≤ ≤ ρ(a, α) f B α∗ .
ρ(a, α) A 1 − |f ◦ φa (z)|2 A
By (2.3) we get
1 (1 − |z|2 )α
f B α∗ ≤ ≤ ρ(a, 1) f B α∗ .
ρ(a, 1) A 1 − |φa ◦ f (z)|2 A
Similar estimations are used when f ∈ B α∗

A,0 .
3. Characterizations for α-Bloch-Bergman classes

We obtain here different characterizations of hyperbolic Bloch Bergman classes.
We imitate the procedure of R. Zhao [Zha].
Define the pseudohyperbolic disk by
D(a, r) = { z ∈ D : |φa (z)| < r }
with Lebesgue measure
(1 − |a|2 )2
(3.1) |D(a, r)| = πr 2
(1 − |a|2 r 2 )2
The following lemmas appear in [Zha]
Lemma 3.1. Let 0 < r < 1. Then for every z ∈ D(a, r),
(1 − |z|2 )2 ∼ |D(a, r)| .
Lemma 3.2. Let c = .0183403 . . .. Then for each c ≤ r ≤ 1
1
ln ≤ 4(1 − r 2 ) < 4.
r
107
5
Lemma 3.3. Let 0 ≤ k1 < ∞, 0 ≤ k2 < ∞ be with k1 − k2 > −1. Then

1
C(k1 , k2 ) = lnk1 (1 − |z|2 )−k2 dx dy < ∞ .
D |z|
Theorem 3.4. Let 0 ≤ α ≤ 1, 0 < R < 1 and 1 < s < ∞. For an analytic
function f ∈ B(D) the following are equivalent:
(A)
(1 − |a|2 )α
sup <∞ .
a∈D 1 − |f (a)|
2
(B)

1 1
sup α dx dy .
a∈D |D(a, R)|1− 2 D(a,R) 1 − |f (z)|2
(C)

(1 − |z|2 )α−2
sup dx dy < ∞.
a∈D D(a,R) 1 − |f (z)|2
(D)

(1 − |z|2 )α−2
sup (1 − |φa (z)|2 )s dx dy < ∞ .
a∈D D 1 − |f (z)|2
(E)

(1 − |z|2 )α−2 s
sup g (z, a) dx dy < ∞ .
a∈D D 1 − |f (z)|2
(F)

|φa (z)|2 1
sup lnα dx dy < ∞ .
a∈D D 1 − |f (z)|2 |z|
Proof. (B) =⇒ (A). By the proof of Corollary 2.2, as − ln(1 − |f (z)|2 ) is
subharmonic and ex is convex −(1 − |f (z)|2 ) results subharmonic, so we have
2π
1 1 1
≤ dθ .
1 − |f (0)|2 2π 0 1 − |f (reiθ )|2
Multiplying by r and integrating from 0 to R < 1 we get

1 1 1
≤ dx dy .
1 − |f (0)|2 πR 2
DR 1 − |f (z)|
2
1
As the function z → is also subharmonic, with the change of vari-
1 − |f (φa (z))|2
able w = φa (z) and by (2.4)

1 1 |φa (w)|2
≤ du dw
1 − |f (a)| 2 πR 2
D(a,R) 1 − |f (w)|
2

1 (1 + |a|)2 1
≤ du dw,
πR2 (1 − |a|)2 D(a,R) 1 − |f (w)|2
therefore

(1 − |a|2 )α (1 + |a|)2 (1 − |a|2 )α 1
≤ du dw .
1 − |f (a)|2 πR2 (1 − |a|)2 D(a,R) 1 − |f (w)|2
108
106
6 J. L
By (3.1), we have

(1 − |a|2 )α 16 1
≤ α α du dw .
1 − |f (a)|2 π 2 Rα (1 − R2 )2−α |D(a, R)|1− 2 D(a,R) 1 − |f (w)|2
By Lemma 3.1 (B) and (C) are equivalent.
(D)=⇒ (C). Since 1 − R2 ≤ 1 − |φa (z)|2 for z ∈ D(a, R), we have for any 1 < s < ∞

(1 − |z|2 )α−2 1 (1 − |z|2 )α−2
dx dy ≤ (1 − |φa (z)|2 )s dx dy.
D(a,R) 1 − |f (z)|
2 (1 − R2 )s D 1 − |f (z)|
2
(E) =⇒ (D). It is a consequence of 1 − |φa (z)|2 ≤ 2g(z, a) for z, a ∈ D .

(A) =⇒ (E) consider the change of variable z = φa (w) in the following integral and
(1.1)

(1 − |z|2 )α−2 s (1 − |φa (w)|2 )α−2 1
g (z, a) dx dy = |φa (w)|2 lns du dv
D 1 − |f (z)| D 1 − |f (φa (w))| |w|
2 2

(1 − |φa (w)|2 )α 1
= (1 − |w|)−2 lns du dv.
D 1 − |f (φa (w))| |w|
2
(3.2)
From Lemma 3.3 and (3.2)

(1 − |z|2 )α−2 s 1
g (z, a) dx dy ≤ f B α∗ (1 − |w|)−2 lns du dv
D 1 − |f (z)| |w|
2 A
D
= f B α∗ C(s, 2) .
A
(A) =⇒ (F ). Since for z ∈ Dr , |φa (z)|2 ≤ (1−r)

1
4 and (1 − |z| )
2 −α
≤ (1−r1 2 )α , then

|φa (z)|2 α 1 1
ln dx dy ≤ f B α∗ |φa (z)|2 (1 − |z|2 )−α lnα dx dy
Dr 1 − |f (z)| 2 |z| A
Dr |z|

1 1
≤ f B α∗ lnα dx dy
(1 − r)4 (1 − r 2 )α A
Dr |z|
C(α, 0)
≤ f B α∗
(1 − r)4 (1 − r 2 )α A
By Lemma 3.3, the left hand is finite. By Lemma 3.2 we have

|φa (z)|2 α 1 α |φa (z)|2
ln dx dy ≤ 4 (1 − |z|2 )α dx dy
D\Dr 1 − |f (z)| |z| D\Dr 1 − |f (z)|
2 2

α
≤ 4 f B α∗ |φa (z)|2 dx dy
A
D\Dr
α
≤ 4 π f B α∗ .
A
1
(F ) =⇒ (D). We know that 1 − |z|2 ≤ 2 ln for z ∈ D, then
|z|

(1 − |z|2 )α−2 (1 − |z|2 )α
(1 − |φa (z)|2 )2 dx dy = |φ (z)|2 dx dy
2 a
D 1 − |f (z)|2 D 1 − |f (z)|

1 1
≤ 2 α
|φ (z)|2 lnα
2 a
dx dy .
D 1 − |f (z)| |z|
109
7
So we have (F ) =⇒ (D) for s = 2. However we see that (D) =⇒ (C) if D is

satisfied for some 1 < s < ∞ and C does not depend of s. Then (C) is equivalent
with (B), (B) =⇒ (A) =⇒ (E) =⇒ (D) for all 1 < s < ∞.
Theorem 3.5. Let 0 ≤ α ≤ 1, 0 < R < 1 and 1 < s < ∞. For an analytic
function f ∈ B(D) the following are equivalent:
(a)
(1 − |z|2 )α
lim =0.
|z|→1− 1 − |f (z)|2
(b)

1 1
lim α dx dy = 0 .
|a|→1− |D(a, R)|1− 2 D(a,R) 1 − |f (z)|2
(c)

(1 − |z|2 )α−2
lim − dx dy = 0.
|a|→1 D(a,R) 1 − |f (z)|2
(d)

(1 − |z|2 )α−2
lim − (1 − |φa (z)|2 )s dx dy = 0 .
|a|→1 D 1 − |f (z)|2
(e)

(1 − |z|2 )α−2 s
lim g (z, a) dx dy = 0 .
|a|→1− D 1 − |f (z)|2
(f )

|φa (z)|2 1
lim lnα dx dy = 0 .
|a|→1− D 1 − |f (z)| 2 |z|
Proof. (a) =⇒ (e). In particular f B α∗ < ∞. Given > 0, take c < R < 1
such that
4s π(1 − R2 )s−1
=.
s−1
By (3.2) we have

(1 − |z|2 )α−2 s (1 − |φa (w)|2 )α 1
g (z, a) dx dy = (1 − |w|)−2 lns du dv
D 1 − |f (z)| DR 1 − |f (φa (w))| |w|
2 2

(1 − |φa (w)|2 )α 1
+ (1 − |w|)−2 lns du dv
D\DR 1 − |f (φ a (w))| 2 |w|
Then by Lemma 3.2

(1 − |φa (w)|2 )α 1
(1 − |w|)−2 lns du dv ≤ s
4 f α∗
BA (1 − |w|2 )s−2 du dv
D\DR 1 − |f (φ a (w))| 2 |w| D\DR
4s π(1 − R )
2 s−1
= f BAα∗ = f BAα∗ .
s−1
Again, by hypothesis there exists 0 < R ≤ R0 < 1 such that for every a ∈ D \ DR0
and w ∈ DR we have
(1 − |φa (w)|2 )α
<.
1 − |f (φa (w))|2
110
108
8 J. L
As in (3.2)

(1 − |φa (w)|2 )α 1 1
(1 − |w|)−2 lns du dv ≤ (1 − |w|2 )−2 lns du dv
DR 1 − |f (φa (w))| |w| |w|
2
DR
≤ C(s, 2).
From these estimations we get (e).
The following implications: (e) ⇒ (d) ⇒ (c) ⇔ (b) ⇒ (a) follow in the same way
as the previous theorem.
(a) =⇒ (f ). Given > 0, there exists 0 < r < 1 such that, for all z ∈ D − Dr
(1 − |z|2 )α
<.
1 − |f (z)|2
By Lemma 3.2 then

|φa (z)|2 1 |φa (z)|2
lnα dx dy ≤ α
4 (1 − |z|2 )α dx dy
D\Dr 1 − |f (z)| 2 |z| D\Dr 1 − |f (z)|2
≤ 4α π .
Let 0 < δ be such that |φa (z)|2 < for all z ∈ Dr and 1 − |a| < δ. Then

|φa (z)|2 α 1 1
ln dx dy ≤ f B α∗ (1 − |z|2 )−α lnα dx dy
Dr 1 − |f (z)| |z| |z|
2 A
Dr
≤ C(α, α) Bα∗ .
A
1
(f) ⇒ (d). For s = 2 it follows from the fact 1 − |z|2 ≤ 2 ln and we repeat the
|z|
similar argument of the previous Theorem.
4. Carleson and series characterization of B α∗ α∗

A and B A,0
In this section we use Carleson type measures to characterize the classes B α∗

A
and B α∗
A,0 . The next definition is due to [Xi].
For 0 < s < ∞, we say that a positive measure µ defined on D is a bounded
s-Carleson measure provided
µ(S(I))
sup <∞
I⊂T |I|s
where |I| denotes the arc lenght of I ⊂ T and S(I) denotes the Carleson box based
on I, that is
z |I|
S(I) = z ∈ D : ∈ I, 1 − |z| ≤ .
|z| 2π
If
µ(S(I))
lim =0
|I|→0 |I|s
for I ⊂ T we say that µ is a compact s-Carleson measure. The following lemma
characterizes s-Carleson measures (see Lemma 2.2 in[AuStXi]).
Lemma 4.1. Let 0 < s < ∞ and µ be a positive measure on D. Then
i) µ is a bounded s-Carleson measure if and only if

sup |φa (z)|s dµ(z) < ∞ ;
a∈D D
111
9
ii) µ is a compact s-Carleson measure if and only if

lim |φa (z)|s dµ(z) = 0 .
|a|→1 D
Then the following result is immediate:

Theorem 4.2. Let 0 < α ≤ 1 and 1 ≤ s < ∞. Let f ∈ B(D) and
(1 − |z|2 )α+s−2
dµ(z) := dx dy .
1 − |f (z)|2
Then
i) f ∈ B α∗
A if and only if dµ(z) is a bounded s-Carleson measure.
ii) f ∈ B α∗
A,0 if and only if dµ(z) is a compact s-Carleson measure.
Proof. It is immediate from Theorems 3.4 (D), 3.5 (d), and Lemma 4.1.
Also, from the Theorems 3.4 and 3.5 we get the following characterization that
involves the behavior of certain series expansions. Define for f ∈ B(D)
i
1 f (z)
(4.1) a{i}
n = dz
2πi γ z n+1
and γ is the circle |z| = 12 .
Theorem 4.3. Let 0 < α ≤ 1. Then f ∈ B α∗ α∗
A (respectively f ∈ B A,0 ) if and
only if
∞
∞
2

2 α {i} n
sup(1 − |z| ) an z < ∞ ,
z∈D
i=0 n=0
(respectively
∞
∞
2

2 α {i} n
lim (1 − |z| ) an z = 0 .)
|z|→1−
i=0 n=0
Proof. Let f ∈ B(D) . Then by (4.1)

∞
∞
2
∞ ∞
1 i {i} n
= |f (z)| =
2i
|f (z)| =
2
an z
1 − |f (z)| 2
i=0 i=0 i=0 n=0
and the theorem follows from this equality.

Theorem 4.4. Let 0 < α ≤ 1 and 1 ≤ s . Then f ∈ B α∗ α∗
A (respectively f ∈ B A,0
) if and only if
∞ ∞ n {i} 2
(1 − |a|2 )s an Γ(n − m + s) n−m
sup
a∈D i=0 n=0 (n + 1)
α+s−1 (n − m)!
a <∞ ,
m=0
(respectively
∞ ∞ n {i} 2
(1 − |a|2 )s an Γ(n − m + s) n−m
lim a =0. )
|a|→1−
i=0 n=0
(n + 1)α+s−1 m=0 (n − m)!
Proof. See [Re To] or follow the idea given in [Xi], p. 23.
112
110
10 J. L
Theorem 4.5. Let 0 ≤ s < ∞. If

(1 − |z|2 )α−2
dx dy < ∞
D 1 − |f (z)|
2
then f ∈ B α∗
A,0
Proof. Let 0 < r < 1 . Then we have

1
(1 − |z|2 )α−2 (1 − |φa (z)|2 )s dx dy
D 1 − |f (z)|2

1 (1 − |z|2 )α−2+s (1 − |a|2 )s
= dx dy
1 − |f (z)|2 |1 − az|2s
Dr
1
+ (1 − |z|2 )α−2 (1 − |φa (z)|2 )s dx dy
D\Dr 1 − |f (z)|2

2 s 1 1
≤ (1 − |a| ) s+2−α
dx dy
Dr 1 − |f (z)| (1 − |z|)
2

1
+ (1 − |z|2 )α−2 dx dy < ∞.
D\Dr 1 − |f (z)|2
Given > 0, by hypothesis and the absolute continuity of the integral, there exists
0 < R < 1 such that the second summand at right side is less than 2 for all
R < r < 1. Also, there exists δ > 0 such that, for all a ∈ D with 0 < 1 − |a| < δ,
the first sumand at right side is less than 2 . So the result follows from Theorem
3.5.
5. Properties of Bα∗
A
Li in [Li] defined for f ∈ B(D)

1
λ(f )(z) = ln .
1 − |f (z)|2
We say that f ∈ Hλ if e2λ(f ) has an harmonic majorant.
It is clear that the sets B α∗ α∗
A or B A,0 are not vector spaces. However they have
some other properties that we will show.
Define for each measurable set, E ⊂ D, 0 < α ≤ 1, 1 ≤ s < ∞

νa,α,s (E) = (1 − |z|2 )α−2 (1 − |φa (z)|2 )s dx dy .
E
Then νa,α,s is absolutely continuous with respect to Lebesgue measure on D and

1 1
dν a,q,s (z) = (1 − |z|2 )α−2 (1 − |φa (z)|2 )s dx dy .
D 1 − |f (z)| D 1 − |f (z)|
2 2
The following theorem results now immediate.

Theorem 5.1. Let 0 < α ≤ 1 and 1 ≤ s < ∞. Then f ∈ B α∗ A (respectively
f ∈ B α∗
A,0 ) if and only if

sup eλ(f )(z) dνa,α,s (z) < ∞
a∈D D
(respectively
lim eλ(f )(z) dνa,α,s (z) = 0 . )
|a|→1− D
113
11
Define for f ∈ B(D)

(1 − |z|2 )α−2
Φ(f, a, t, α) = dx dy
D(a,t) 1 − |f (z)|2
where 0 < t < 1, and D(a, t) = { z ∈ D : |φa (z)| < t} is the pseudohyperbolic
disk.
Theorem 5.2. Let f ∈ B(D), 0 < α ≤ 1 and 0 ≤ s < ∞ . Then f ∈ B α∗
A (or
B α∗
A,0 ) if and only if
1
sup Φ(f, a, t, α)(1 − t)s−1 dt < ∞,
a∈D 0
(or
1
lim − Φ(f, a, t, α)(1 − t)s−1 dt = 0. )
|a|→1 0
Proof. By definition of Φ(f, a, t, α) and Fubini’s Theorem we have

1
1
(1 − |z|2 )α−2
Φ(f, a, t, α)(1 − t)s−1 dt = dx dy (1 − t)s−1 dt
D(a,t) 1 − |f (z)|
2
0 0

(1 − |z|2 )α−2 1
= (1 − t)s−1 dt dx dy .
D 1 − |f (z)|
2
|φa (t)|
Since
1
1 1 (1 − |φa (z)|2 )s
(1 − t)s−1 dt = (1 − |φa (z)|)s =
|φa (t)| s+1 s + 1 (1 + |φa (z)|)s
then
1
1 dνa,α,s (z) 1 dνa,α,s (z)
≤ Φ(f, a, t, α)(1 − t)s−1 dt ≤
2s (s + 1) D 1 − |f (z)|2 0 s+1 D 1 − |f (z)|2
and the theorem follows from the previous estimation and Theorem 3.1.
Theorem 5.3. Let 0 < α ≤ 1 and 0 ≤ s < ∞ . Then
B(D) · B α∗ α∗ α∗
A = B A · B(D) ⊂ B A
and
B(D) · B α∗ α∗ α∗
A,0 = B A,0 · B(D) ⊂ B A,0 .
Proof. Let f ∈ Bα∗

A and g ∈ B(D) . Then |f (z)g(z)| ≤ |f (z)|. Therefore
(1 − |z|2 )α (1 − |z|2 )α
≤
1 − |f (z)g(z)|2 1 − |f (z)|2
and the theorem follows from this estimation.
We prove now that the classes are convex.
Theorem 5.4. Let 0 < α ≤ 1 and 0 ≤ s < ∞. Then for all 0 ≤ t ≤ 1
(1 − t) · B α∗ α∗ α∗
A + tB A = B A
and
(1 − t) · Bα∗ α∗ α∗
A,0 + tB A,0 = B A,0 .
114
112
12 J. L
Proof. Let f ∈ B α∗
A and g ∈ B(D). Define
Ω = { z ∈ D : |f (z)| ≤ |g(z)| } .
Then for z ∈ Ω
|(1 − t)f (z) + tg(z)| ≤ (1 − t)|f (z)| + t|g(z)| ≤ (1 − t)|g(z)| + t|g(z)| = |g(z)| .
Similarly if z ∈ D \ Ω then |(1 − t)f (z) + tg(z)| ≤ |f (z)|. Therefore
(1 − |z|2 )α (1 − |z|2 )α (1 − |z|2 )α
≤ +
1 − |(1 − t)f (z) + tg(z)|2 1 − |g(z)| 2 1 − |f (z)|2
and the theorem follows from this estimation.
For f, g ∈ B ∗α
A define

f (z)
2 α g(z)
(5.1) d(f, g) = sup(1 − |z| ) − .
z∈D 1 − |f (z)|2 1 − |g(z)|2
Observe that
d(f, 0) = f B ∗α .
A
We now prove that d is a distance function on B α∗

A .
Theorem 5.5. Let 0 < α ≤ 1. Then (B∗α

A , d) is a convex complete metric
space.
Proof. It sufficies to prove that d(f, g) = 0 implies f ≡ g, for f, g ∈ B ∗α
A . By
definition of the metric d
f (z) g(z)
(5.2) =
1 − |f (z)|2 1 − |g(z)|2
for all z ∈ D. In particular
|f (z)| |g(z)|
= .
1 − |f (z)|2 1 − |g(z)|2
t
Now consider the function h : [0, 1) → R given by h(t) = . Then h (t) =
1 − t2
t2 + 1
> 0 and therefore h(t) is an increasing function. From here h(|f (z)|) =
(1 − t2 )2
h(|g(z)|) if and only if |f (z)| = |g(z)| for all z ∈ D. It is clear that if g ≡ 0 then
f (z)
f ≡ 0. In other case the function z → is defined almost everywhere in D and
g(z)
f (z)
is analytic in its domain. As = 1 for all z in its domain, then f (z) = c with
g(z) g(z)
|c| = 1. Therefore by (5.2) we get again f ≡ g.
We prove now that d is complete. Let {fn } ⊂ B ∗α A be a Cauchy sequence. Then
given > 0 there exists N ∈ N such that

2 α
fn (z) fm (z)
d(fn , fm ) = sup(1 − |z| ) − <
z∈D 1 − |fn (z)|2 1 − |fm (z)|2
for all n ≥ m ≥ N . It is clear that {fn } is totally bounded, so its a normal family
and then we can suppose that {fn } converges to f . In particular

2 α
fn (z) fm (z)
(1 − |z| ) − <
1 − |fn (z)|2 1 − |fm (z)|2
115
13
and taking limit when n → ∞ (and m fixed) we obtain

f (z)
2 α fm (z)
(5.3) (1 − |z| ) − <.
1 − |f (z)|2 1 − |fm (z)|2
From (5.3) we get that |f (z)| < 1 for all z ∈ D and moreover
|f (z)| |fm (z)|
(5.4) (1 − |z|2 )α ≤ (1 − |z|2 )α + .
1 − |f (z)| 2 1 − |fm (z)|2
It follows that
f B ∗α ≤ fm B ∗α + .
A A
Then f ∈ B ∗α
A and by (5.3), d(fm , f ) → 0 if m → ∞.
Corollary 5.6. Let 0 < α ≤ 1. Then B∗α
A,0 is a closed (complete) subspace of
B ∗α
A .
Proof. Let {fn } be a convergent sequence in B ∗α

A,0 . Then the result it follows
directly from the estimation (5.4).
Example 5.7. For a ∈ D and 0 < α < 1, it follows immediately from (1.1)
/ B ∗α
that φa ∈ ∗
/ B ∗a,0 .
a and φa ∈ B A but φa ∈
n
Example 5.8. Let fn : D → D given by fn (z) = . Then fn ∈ B(D) and
n+1
{fn } is a sequence that converges uniformly by compact sets on D. However {fn }
is not a Cauchy sequence in B ∗α
A since for n ≥ m, we have

n(n + 1) m(m + 1)
d(fn , fm ) = − sup(1 − |z|2 )α
2n + 1 2m + 1 z∈D
therefore if m is fixed and n → ∞ then d(fn , fm ) → ∞.

Example 5.9. Let f ∈ B(D) be such that f (D) ⊂ D then f ∈ B ∗α
A,0 .
Proposition 5.10. Let 0 < α ≤ 1. Then

B ∗α
A ⊂ B ∗γ
A,0
α<γ
Proof. Let f ∈ B ∗α
A . Then for any 0 < t < ∞ we have
(1 − |z|2 )α+t (1 − |z|2 )α
0 ≤ lim − ≤ lim − (1 − |z|2 )t
|z|→1 1 − |f (z)|2 |z|→1 1 − |f (z)|2
≤ f B ∗α lim (1 − |z|2 )t = 0 .
A |z|→1−
For 0 < β < 1 consider the function fβ : D → D defined by fβ (z) = 1 − (1 − z)β

(see [Li]). Then
|fβ (z)|2 = 1 − 2Re (1 − z)β + |1 − z|2β .
Let 0 < ρ < 1 and set z = 1 + ρeiθ for π
2
<θ< 3π
2
, such that, cos θ < − ρ2 , then
(1 − z)β = (−ρeiθ )β = ρβ ei(θ+π)β .
Therefore
1 − |fβ (z)|2 = 2Re (1 − z)β − |1 − z|2β = 2Re ρβ ei(θ+π)β − ρ2β
= 2ρβ cos(θ + π)β − ρ2β
116
114
14 J. L
and
1 − |z|2 = ρ(−2 cos θ − ρ).
It is clear that if z ∈ D − Dρ (1) then
(1 − |z|2 )α (1 − |z|2 )α
lim − = lim =0.
|z|→1 1 − |fβ (z)|2 |z|→1− 2Re (1 − z)α − |1 − z|2α
We study now the case when z ∈ D ∩ Dρ (1). Since
(1 − |z|2 )α ρα (−2 cos θ − ρ)α ρα (−2 cos θ − ρ)α

= β = β
1 − |fβ (z)| 2 2ρ cos(θ + π)β − ρ 2β ρ (2 cos(θ + π)β − ρβ )
∗β ∗β
then fβ ∈ B ∗α / B ∗α
A,0 if β < α, fβ ∈ B A − B A,0 , and fβ ∈
∗α
A,0 , B A if β > α.
From our previous estimations we have:
Theorem 5.11. Let 0 < α < β ≤ 1 . Then

∗β ∗β
B ∗α ∗α
A,0 ⊂ B A , B ∗α
A,0 ⊂ B A,0 , and B ∗α
A ⊂ BA .
Moreover
B ∗α
A,0 ⊂ B ∗γ
A,0 .
α<γ
All the inclusions are strict.
Proof. For the first inclusion consider fα . For the second inclusion let α < γ <
∗γ ∗β
β. Then B ∗α
A,0 ⊂ B A ⊂ B A,0 and consider fγ . For the third consider fβ and for the
last fα .
References
[AuReTo] R. Aulaskari, L.F. Reséndis O, L. M. Tovar S. Hyperbolic Bergman classes, Advances
in Applied Clifford Algebras, First on line, 2008.
[AuStXi] R. Aulaskari, D. Stegenga and J. Xiao.,Some subclasses of BM OA and their character-
ization in terms of Carleson measures, Rocky Mountain J. Math. 26, p.485-506.
[AuXiZh] R. Aulaskari, J. Xiao and R. Zhao On subspaces and subsets of BMOA and UBC,
Analysis Vol. 15 (1995), 101–121.
[Co] J. B. Conway Functions of One Complex Variable, Graduate Texts in Mathematics, 1973,
(1st ed.)
[Re To] L. F. Reséndis O., L.M. Tovar S. On hyperbolic clases of analytic functions, Bull. Sci.
Lettres de Lodz (Poland) 58, Ser. Rech. Deform. 55, (2008).
[Xi] J. Xiao, Holomorphic Q Classes, Lectures Notes in Mathematics, Springer (2001), # 1767.
[Li] X. Li, On Hyperbolic Q Classes, Mathematica, Dissertationes, Ann. Acad. Scie. Fenn., # 145
(2005)
[Ya] Yamashita, Hyperbolic Hardy Class H 1 , Math. Scand 45 (1979) pp. 261-266.
[Ya1] Yamashita, A non-normal function whose derivative has finite area integral of order 0 <
p < 2, Ann. Acad. Scie. Fenn. Series A. I. Mathematica Volume, (1978-1979) pp. 293-298.
[Ya2] Yamashita, A non-normal function whose derivative has finite area integral of order 0 <
p < 2, Ann. Acad. Scie. Fenn. Series A. I. Mathematica Volume, (1982) pp. 349-365.
[Zha] R. Zhao, On α-Bloch functions and V M OA., Acta Mathematica Scientia, 1996, 16 (3):
349-360.
[Zha1] R. Zhao, On a general family of functions spaces, Ann. Acad. Scie. Fenn. Math. Diss, 105.
[Zh] K. Zhu, Operator Theory in Function Spaces, MARCEL DEKKER, INC. 1990. New York.
117
15
(J. L
awrynowicz) University of Lodz, Institute of Physics and Polish Academy of
Science, Pomorska 149/153, PL-90-236 Lodz, Poland
E-mail address: jlawryno@uni.lodz.pl
(L. F. Reséndis O.) Universidad Autónoma Metropolitana, Unidad Azcapotzalco,

C.B.I. Apartado Postal 16-306 C.P. 02200 México 16, D.F. Area de Análisis Matemático
y sus Aplicaciones
E-mail address: lfro@correo.azc.uam.mx
Escuela Superior de Fı́sica y Matemáticas del IPN. Edif. 9, Unidad ALM, Zacatenco
del IPN., C.P. 07300, D.F., México
E-mail address: tovar@esfm.ipn.mx
Volume 509, 2010
Some words about the application of Tchebycheff systems to

Weak Hilbert’s 16th Problem
J. Tomás Lázaro
Abstract. In this talk we will try to introduce (in a very naı̈ve way) the
so-called Weak Hilbert’s Problem, posed by Arnol’d in 1977, its relation with
the original Hilbert’s 16th Problem and how Tchebycheff systems have been
applied to approach them.
Contents
1. Hilbert’s 16th and Weak Hilbert’s 16th Problems 119
2. Estimating the number of zeroes of an Abelian integral 123
References 127
1. Hilbert’s 16th and Weak Hilbert’s 16th Problems

The aim of this lecture is to introduce the reader to the so-called Weak Hilbert’s
16th Problem and to show how Tchebycheff systems can be used to give some
approximative and partial solution to some concrete systems. It is addressed to
a general audience with no expertise, in principle, on this topic and almost all
the results contained are known. Thus we will spend an important part of this
notes on introducing the problem, listing some of the main references and showing
properties of the systems we will deal with but without entering into details. For
more information we refer the reader to some references like, for instance, [12,
14] - which have been our main sources preparing this short lecture- , but the
amount of papers written on this topic is huge as well as the number of researchers
involved (Arnol’d, Bautin, Chen, Chicone, Christopher, Chow, Dulac, Dumortier,
1991 Mathematics Subject Classification. Primary 34C07; Secondary 34C08.

Key words and phrases. Weak Hilbert’s 16th Problem, limit cycles, Abelian integrals,
Tchebycheff systems .
The author is supported by the Ministerio de Educación y Ciencia grant number MTM2006-
00478. He also thanks Centre de Recerca Matemàtica for the hospitality and facilities during the
preparation of these notes.
2010
c c Mathematical
American
0000 (copyright Society
holder)
1
119
117
120
118
2 J. TOMÁS LÁZARO
Écalle, Françoise, Gavrilov, Gasull, Iliev, Ilyashenko, Li, Llibre, Lloyd, Mardesic,
Petrov, Poincaré, Pontryagin, Roussarie, Rousseau, Schlomiuk, Yakovenko, Zhang,
Zoladek, .... and many others that we have not included in this short list (we
apologize for that)). Let us now to introduce the framework of this lecture.
One of most exciting problems contained in Hilbert’s celebrated list (see [10]),
devoted to the so-called Problems of the XXth Century, is 16th Problem whose first
part reads, in original version, approximatively like this: Which is the number and
relative position of Poincaré limit cycles (isolated periodic orbits) that can have a
polynomial differential equation
dy Pn (x, y)
(1.1) = ,
dx Qn (x, y)
where Pn , Qn are polynomials of degree n ?
Notice that the problem is trivial for n = 1 since a linear system cannot have
limit cycles, so we will assume always that n ≥ 2. Following [12], this problem
admits at least three different specifications, which are
(1) Individual finiteness problem: Given a polynomial differential equation (1.1),
with deg Pn , Qn ≤ n, to prove that it has only a finite number of limit
cycles.
Dulac claimed in 1923 that he had solved it but the proof he gave was
erroneous. Much later, Ilyashenko in 1991 [11] and Écalle in 1992 [6] gave
two different proofs of Dulac’s Theorem.
(2) Existential Hilbert Problem: To prove that, for any n ∈ N, the number
of limit cycles of all polynomial differential equations (1.1) of degree less
or equal than n is uniformly bounded. This uniform bound is denoted by
H(n) so this problem can be rewritten as to prove that H(n) < ∞ for any
n ∈ N.
(3) Constructive Hilbert Problem: To give an upper bound for H(n) or to
suggest an algorithm to get it.
It is important to stress that, of these three problems only the first one (the weakest)
has been solved. A remarkable number of papers has been published concerning
such problems providing partial solutions but no proof has been obtained for the
general statements.
Regarding the second part of Hilbert’s problem, the one concerning the configu-
ration of limit cycles, we want just to mention the paper of Llibre and Rodrı́guez [17],
where a general result was obtained.
Let us now introduce a weaker version on Hilbert’s 16th Problem. It corre-
sponds to a Problem posed by Arnol’d in 1977 and that can be stated as follows:
Let H = H(x, y) a polynomial in x, y of degree m ≥ 2 and assume that the
level curves γh ⊂ {(x, y)| H(x, y) = h} form a continuous family of ovals {γh } for
h1 < h < h2 . Consider now a polynomial 1-form ω = f (x, y) dy − g(x, y) dx with
f, g polynomials satisfying that max {deg(f ), deg(g)} = n ≥ 2. Then, the problem
consists on, for fixed integer values m and n, finding the maximum number Z(m, n)
of isolated zeroes of the Abelian integrals

I(h) = ω.
γh
We recall that an Abelian integral is the integral of a rational 1-form along an

algebraic oval. Following [14], we will show that this problem of counting the
TCHEBYCHEFF SYSTEMS AND WEAK HILBERT’S 16TH PROBLEM 121
119
3
number of zeroes of an Abelian integral is closely related to Hilbert’s 16th Problem.

Let us see it.
Consider H(x, y) a polynomial of degree m ≥ 2 and the corresponding Hamil-
tonian vector field
ẋ = −Hy (x, y)
ẏ = Hx (x, y)
where ˙ denotes d/dt and Hu stands for the partial derivative ∂H/∂u, for u = x, y.
Let us introduce ε an small parameter and consider the following perturbation of
the previous system

ẋ = −Hy (x, y) + εf (x, y)
(1.2)
ẏ = Hx (x, y) + εg(x, y).
For the unperturbed system (ε = 0), let us suppose that the origin is a center
equilibrium point and that, therefore, for energies h1 < h < h2 there exists a
family of ovals {γh } varying continuously with h and filling up an annulus around
it for h ∈ (h1 , h2 ). The corresponding Abelian integral reads in this case as

(1.3) I(h) = f (x, y) dy − g(x, y) dx.
γh
It is known that when we perturb such a system, i.e. take ε = 0, some of the initial
periodic orbits γh (slightly deformed) will persist and will remain isolated. That is,
they will become limit cycles of the unperturbed system. Indeed, the perturbation
of a system having a linear center at the origin (that is, eigenvalues of its differential
at the origin are purely imaginary conjugated complex) is a classical way to produce
limit cycles.
A natural question which arises is the following: Is there any value h ∈ (h1 , h2 )
and some (isolated) periodic orbits Γε of the perturbed system such that Γε tend
to γh as ε → 0? How many of such Γε can we have for the same h? If this happens
we will say that Γε bifurcates from γh .
The usual way to approach to this problem is to consider Σ, a transversal section
(a segment in this case) to the periodic orbit γh , and to choose H to parameterize
such a section Σ. We can denote by γε (h) the orbit of the perturbed system (1.2),
close to γh provided ε small enough, with starting point h on Σ. We define P (h, ε)
the next intersection of γε (h) with Σ (see Fig. 1) and consider
d(h, ε) = P (h, ε) − h
the so-called displacement function. Notice that (isolated) zeroes of d(h, ε) corre-
spond to limit cycles of system (1.2). The following Theorem states a relationship
between this displacement function and the Abelian integral (1.3).
Theorem 1.1 (Poincaré-Pontryagin [21, 22]). Under the hypotheses above one
has that
d(h, ε) = εI(h) + ε2 φ(h, ε) as ε → 0,
φ(h, ε) being analytic and uniformly bounded in a compact region containing (h, 0)
for h1 < h < h2 .
More precisely, next result provides a direct relation between zeroes of the
Abelian integral I(h) and existence of limit cycles.
Theorem 1.2. Provided the Abelian integral I(h) in (1.3) is not identically
zero for h1 < h < h2 , the following statements hold:
122
120
γε (h)
P (h, ε)
γh
Σ
h
Figure 1. Displacement function
• If system (1.2) has a limit cycle bifurcating from γh∗ then I(h∗ ) = 0.
• If there exists h1 < h∗ < h2 such that I(h∗ ) = 0 and I (h∗ ) = 0 then
system (1.2) has a unique limit cycle bifurcating from γh∗ . Moreover, this
limit cycle is hyperbolic.
• If h∗ is a zero of order k of I(h), i.e. I(h∗ ) = I (h∗ ) = . . . = I (k−1) (h∗ ) =
0 and I (k) (h∗ ) = 0, then system (1.2) has at most k limit cycles which
bifurcate from γh∗ (taking into account the multiplicity of the limit cycles).
• Finally, the total number (taking into account
multiplicity) of limit cycles
bifurcating from the period annulus γh associated to system (1.2)
h1 <h<h2
is bounded by the maximum number of isolated zeroes (counting again
multiplicity) of the corresponding Abelian integral I(h) for h1 < h < h2 .
A classical (and simple) example of application of this theorem is the following
one.
Example: Let us consider the well known Van der Pol equation

ẍ + ε x2 − 1 ẋ + x = 0
or, equivalently

ẋ = y .
(1.4)
ẏ = −x + ε 1 − x2 y
The unperturbed system (ε = 0) is Hamiltonian, with Hamilton function H(x, y) =
x2 + y 2 and the origin as an elliptic equilibrium point. It presents the following
family of ovals {γh }h defined through

γh = (x, y)| x2 + y 2 = h
defined for h > 0. It is straightforward to check, using polar coordinates, that in
this case
2π
h2
I(h) = − (1 − x2 )y dx = 1 − h2 cos2 θ h2 −sin2 θ dθ = πh2 −1 .
γh 0 4
Notice that there is only one positive zero of I(h) (the case h = 0 has not to
be considered since it corresponds to the singularity) is h = 2. It is not difficult
121
5
to check that I (2) = 4π = 0 so, applying the previous theorem, we can deduce
the existence of a unique limit cycle of system (1.4) for small enough values of ε.
Moreover this limit cycle is hyperbolic.
Even though there exists an strong relation between the problem of counting the
maximum number of zeroes of an Abelian integral I(h) and the problem concerning
the number of limit cycles appearing (locally) around an equilibrium, a recent result
by Dumortier, Panazzolo and Roussarie [5] (pointed out to us by the referee) has
shown that this relationship is not completely exact.
2. Estimating the number of zeroes of an Abelian integral

There exist different well known methods to approach the problem of counting
(or giving bounds) the number of zeroes that a given Abelian integral I(h) can
have in its domain of definition. Among them, the most common are those based
on the Picard-Fuchs equation, on the so-called Argument Principle and on the Av-
eraging equation. The first one consists on seeking for and studying the differential
equation satisfied by the basis of functions forming I(h), that is, those functions
I1 (h), I2 (h), . . . , In (h) such that
I(h) = p1 (h)I1 (h) + p2 (h)I2 (h) + · · · + pn (h)In (h)
with p1 (h), p2 (h), . . . , pn (h) polynomials in h. The second one type of methods
deals with Complex Analysis skills, like the Argument Principle itself and Rouché’s
Theorem, which allow us to estimate the number of zeroes that an analytic extension
of I(h) has in a suitable complex domain. Finally, the third one consists on deriving
properties (if possible) of the original system from those of the system obtained after
one or more steps of averaging.
Another type of methods, those we want to present with some more (non really
much more) detail, are those based in the so-called Tchebycheff systems. In a few
words, we would like to find functions J1 (h), J2 (h), . . . , JN (h) such that the Abelian
integral I(h) associated to system (1.2) could be expressed in the form
I(h) = λ1 J1 (h) + λ2 J2 (h) + · · · + λN JN (h)
and where J1 (h), J2 (h), . . . , JN (h) constituted a so-called Tchebycheff system on
its domain of definition, a set of functions with properties similar to those of poly-
nomials. Among many papers using this method to obtain bounds on the number
of zeroes of some Abelian integrals we would stress the works of Petrov [19, 20]
and Mardesic [18].
Before introducing an example of such use, let us remind its definition and list
some of their most relevant properties.
2.1. Tchebycheff systems. Let us consider {g0 , g2 , . . . , gn } a system of con-

tinuous functions defined on a Hausdorff space A. We call it a Tchebycheff system,
T-system in short, if they satisfy the so-called Haar condition, i.e., for any n + 1
distinct points x0 , x1 , . . . , xn of A the vectors
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
g0 (x0 ) g1 (x0 ) gn (x0 )
⎜ g0 (x1 ) ⎟ ⎜ g1 (x1 ) ⎟ ⎜ gn (x1 ) ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ .. ⎟, ⎜ .. ⎟, ... ,⎜ .. ⎟,
⎝ . ⎠ ⎝ . ⎠ ⎝ . ⎠
g0 (xn ) g1 (xn ) gn (xn )
124
122
are linearly independent. In other words, the determinant

g0 (x0 ) g1 (x0 ) · · · gn (x0 )

g0 (x1 ) g1 (x1 ) · · · gn (x1 )

(2.1) .. .. .. .. = 0
. . . .

g0 (xn ) g1 (xn ) · · · gn (xn )
for any n + 1 distinct points x0 , x1 , x2 , . . . , xn in A. The n + 1-dimensional subspace
of C(A) defined by {g0 , g1 , . . . , gn }, denoted by span {g0 , g1 , . . . , gn }, is called a
Tchebycheff space.
This kind of spaces generalizes the behavior of the space of polynomials of
degree less or equal than n, span {1, x, . . . , xn } regarding the number of their zeroes
and the uniqueness of solution of the interpolation problem. Precisely, we have
(see [1, page 92]):
Proposition 2.1. Let g0 , g1 , . . . , gn be n+1 real (or complex) valued continuous
functions defined on a Hausdorff space A containing, at least, n + 1 points. Then
the following assertions are equivalent:
(i) Any non-zero f belonging to span {g0 , g1 , . . . , gn }, the subspace generated
by g0 , g1 , . . . , gn , has at most n distinct zeros in A. Multiple zeroes must
be considered with their multiplicity, that is, a double zero counts like two
zeroes and so on.
(ii) The interpolation problem has a unique solution: for any n + 1 distinct
points x0 , x1 , . . . , xn in A and real (or complex) numbers y0 , y1 , . . . , yn ,
there exists a unique f ∈ span {g0 , g1 , . . . , gn } satisfying that
f (xj ) = yj , j = 0, 1, . . . , n.
(iii) If x0 , x1 , . . . , xn are distinct points of A, then the determinant

g0 (x0 ) · · · gn (x0 )

.. .. ..
. . . = 0

g0 (xn ) · · · gn (xn )
Despite there are no many examples of T-systems, one can find interesting cases
in the literature.
Lemma 2.2. The following sets of functions are T-systems:

(1) xλ0 , xλ1 , . . . , xλn for x ∈ [0, ∞), provided that 0 ≤ λ0 < λ1 < λ2 <
. . . < λn . It is also a T-system on (0, ∞) if λ0< λ1 < λ2 < . . . < λn .

(2) xλ0 , xλ0 log x, xλ1 , xλ1 log x, . . . , xλn , xλn log x in (0, ∞) provided λ0 <
λ1 < λ 2 < . . . < λ n .
(3) If λ0 < λ1 < λ2 < . . . < λn , the system

1 1 1
, ,..., ,
x − λ0 x − λ1 x − λn
for x ∈ R \ {λ0 , λ1 , . . . , λn }.
(4) {1, cos x, sin x, cos 2x, sin 2x, . . . , cos nx, sin nx} for x ∈ [0, 2π) and {1, cos x,
cos 2x, . . . , cos nx} for x ∈ [0, π).
On the other hand, the following useful properties satisfied by T-systems are
known:
123
7
Lemma 2.3. Let A be a Hausdorff space with at least n + 1 distinct points.

Then:
(i) If {g0 , g1 , . . . , gn } is a T-system on A then it is also a T-system on any
subset B ⊂ A, provided B contains at least n + 1 distinct points.
(ii) If {g0 , . . . , gn } is a T-system on A and f ∈ C(A) has constant sign on A,
then {f g0 , f g1 , . . . , f gn } is also a T-system on A.
(iii) If {g0 , g1 , . . . , gn } is a T-system on [a, b] then
x x x
1, g0 (t) dt, g1 (t) dt, . . . , gn (t) dt,
a a a
is also a T-system on [a, b], provided all gj (x) are Lebesgue integrable on
[a, b].
(iv) The same result holds for
x x x
1, c0 + g0 (t) dt, c1 + g1 (t) dt, . . . , cn + gn (t) dt,
a a a
c0 , c1 , . . . , cn being arbitrary real constants.
For more details on T-systems we refer the reader, for instance, to the books
of Karlin and Studden [13], Cheney [3], Borwein and Erdélyi [1] and references
therein.
2.2. An example. The problem of determining the number of limit cycles

bifurcating from the period annulus of systems of type

ẋ = −yF (x, y) + ε P (x, y)
,
ẏ = xF (x, y) + ε Q(x, y)
for ε small, P (x, y), Q(x, y) polynomials and F satisfying F (0, 0) = 0, has been
widely studied (see, for instance, [15, 24, 25, 4, 16, 9, 2, 8]). In this example we
will consider systems of the form

ẋ = −yFK (x) + ε Pn (x, y)
(2.2) ,
ẏ = xFK (x) + ε Qn (x, y)
Pn (x, y), Qn (x, y) being polynomials of degree n and FK (x) consisting on a set of
vertical straight lines x = aj , j = 1, . . . , K, that is,
K

(2.3) FK (x) = (x − aj ),
j=1
where {a1 , a2 , . . . , aK } are distinct positive real numbers. By construction, any line
x = aj becomes an invariant set of singular points.
A possible application of this method is to look for estimates on the number of
limit cycles which can appear from bifurcations of periodic orbits of the unperturbed
system (ε = 0) covering the period annulus

D = (x, y) ∈ R2 | 0 < x2 + y 2 < a = min aj .
j
This type of problems has been studied in many papers presenting different choices
for the invariant sets of singulars points (isolated points, straight lines with different
multiplicities, conics, etc) [24, 25, 2, 8].
126
124
As it is standard in this kind of problems, we can rewrite our system (2.2) in

D in the equivalent form
⎧ K
⎪
⎪
⎪ −y (x − aj )
⎪
⎨
ẋ = + ε P n (x, y)/
j=1
(2.4) K .
⎪
⎪
⎪ ẏ =
⎪ x + ε Q (x, y)/ (x − a )
⎩ n j
j=1
and to denote by

γr = (x, y) | x2 + y 2 = r 2 , for 0 < r < a1
any periodic orbit in this period annulus of the unperturbed system. As it was men-
tioned in the introduction, isolated zeroes of the displacement map are intimately
related to the number of limit cycles bifurcating from the family of ovals {γr } and,
in particular, to the maximum number of zeroes of the associated Abelian integral

Q(x, y) dx − P (x, y) dy
(2.5) I(r) = .
γr FK (x, y)
One of the aims in the paper [7] was to provide an upper bound for the number
of zeroes of this Abelian Integral associated to a system (2.2), depending on the
number K of critical straight lines and the degree n of the perturbation polynomials
Pn and Qn .
Thus, in [7] it is proved the following result:
Theorem 2.4. Let us consider a system of the form (2.2),

ẋ = −yFK (x) + ε Pn (x, y)
,
ẏ = xFK (x) + ε Qn (x, y)
where
K

FK (x) = (x − aj ),
j=1
with real positive numbers {a1 , a2 , . . . , aK }, are vertical straight lines of singular
points and Pn , Qn are polynomials in x, y of degree n. ε is an small parameter.
For such a system, let us consider the associated Abelian Integral

Qn (x, y) dx − Pn (x, y) dy
I(r) = K
.
γr
(x − aj )
j=1
Then, for K ≥ 1, n ≥ 1, we have that the number of zeroes of I(h), denoted by

Z(I), can be bounded by
n
(2.6) Z(I) ≤ (K + 1) + 1 + 1.
2
Therefore, since Z(I) is an upper bound of ZS (I), the number of simple zeroes
of I, one can deduce that, for ε = 0 small enough, the number of limit cycles
bifurcating from periodic orbits of the unperturbed system is bounded by
n
Z(I) ≤ (K + 1) + 1 + 1.
2
Have in mind, however, that this bound is not optimal.
125
9
The proof starts checking that our Abelian integral I(r) admits an expression
of the form
K ϕ(j) 2
Qn (x, y) dx − Pn (x, y) dy (0) [(n+1)/2] (r )
I(r) = K
= ϕ (r 2
) + ,
γr [n/2]
j=1 a2j − r 2
(x − aj )
j=1
(m)
where ϕs (ρ) are polynomial of degree s in ρ and [z] denotes the integer part of z.
After that (with some effort) one can prove that these functions form a Tchebycheff
vector space and applying results in subsection 2.1 the bound (2.6) is derived.
Acknowledgements I would like to thank the organizers of this Jairo Charris

Seminar, professors Primitivo Acosta-Humanez, Reinaldo Núñez and Jesús Her-
nando Pérez (Pelusa), and the Escuela de Matemáticas of the Universidad Sergio
Arboleda (Bogotá, Colombia) for their support, kindness and hospitality during the
period I spent there. I would like also to thank the referee for pointing out to me
some important points in the previous version.
References
[1] P. Borwein and T. Erdélyi. Polynomials and polynomial inequalities. Graduate Texts in Math-
ematics, Springer, 1995.
[2] A. Buică and J. Llibre. Limit cycles of a perturbed cubic polynomial differential center. Chaos
Solitons Fractals 32, 1059-1069, 2007.
[3] E.W. Cheney. Introduction to Approximation Theory. AMS Chelsea Publishing, First Edition
1966. Second Edition 1982.
[4] B. Coll, A. Gasull and R. Prohens. Bifurcation of limit cycles from two families of centers.
Dyn. Contin. Discrete Impul. Syst. Ser. A math. Anal. 12, 275–287, 2005.
[5] F. Dumortier, D. Panazzolo and R. Roussarie. More limit cycles than expected in Liénard
Equations. Proc. Amer. Math. Soc. 135, number 6, 1895–1904, 2007.
[6] J. Écalle. Introduction aux fonctions analysables and preuve constructive de la conjecture de
Dulac. Hermann, Paris, 1992.
[7] A. Gasull, J. Tomás Lázaro and J. Torregrosa. Preprint. In preparation, 2008.
[8] A. Gasull, R. Prohens and J. Torregrosa. Bifurcation of limit cycles from a polynomial non-
global center. To be published in J. Dyn. Diff. Equat.,2008.
[9] J. Giné and J. Llibre. Limit cycles of cubic polynomial vector fields via the averaging theory.
Nonlinear Anal. 66, 1707–1721, 2007.
[10] D. Hilbert. Mathematical Problems. Transl. Bull. Amer. Math. Soc., 8, 437–479, M. Newton,
1902. Reprinted in Bull. Amer. Math. Soc. 37, 4007–436, 2000.
[11] Yu. Ilyashenko. Finiteness theorems for limit cycles. Amer. Math. Soc., Providence, RI, 1991.
[12] Yu. Ilyashenko and S. Yakovenko, editors. Concerning the Hilbert 16th Problem. Transl.
Amer. Math. Soc., Series 2, 165, Providence, RI, 1995.
[13] S.J. Karlin and W.J. Studden. T-systems: with applications in analysis and statistics. Pure
and Applied Mathematics, Interscience Publishers, 1966.
[14] C. Li. Abelian integrals and aplication to weak Hilbert’s 16th Problem. Advanced course on
limit cycles of differential equations, Quaderns del Centre de Recerca Matemàtica CRM, num.
38, Bellaterra, 2006.
[15] C. Li, J. Llibre and Z. Zhang. Weak focus, limit cycles and bifurcations for bounded quadratic
systems. J. Difer. Equations 115, 193–223, 1995.
[16] J. Llibre, J.S. Pérez del Rı́o and J.A. Rodrı́guez. Averaging analysis of a perturbed quadratic
center. Nonlinear Anal. 46, 45–51, 2001.
128
126
[17] J. Llibre and G. Rodrı́guez. Configuration of limit cycles and planar polynomial vector fields.
J. Diff. Eq. 198, 374–380, 2004.
[18] P. Mardesic. Chebyshev systems and the versal unfolding of the cusps of order n. Travaux en
Cours, 57, Hermann, Paris,1998. xiv+153 pp.
[19] G.S. Petrov. Elliptic integrals and their nonoscillation. Funct. Anal. Appl. 20, No. 1, 46–49,
1986. English transl. Funct. Anal. Appl. 20, No. 1, 37–40,1986.
[20] G.S. Petrov. The Chebyschev property of elliptic integrals. Funct. Anal. Appl. 22, No. 1,
83–84, 1986. English transl. Funct. Anal. Appl. 22, No. 1, 72–73,1988.
[21] H. Poincaré. Sur le problème des trois corps et les équations de la dynamique. Acta Math.,
XIII, 1–270, 1890.
[22] L. Pontryagin. On dynamical systems close to hamiltonian ones. Zh. Exp. & Theor. Phys. ,
4, 234–238, 1934.
[23] O. Shisha. T-systems and best partial bases. Pacific Journal of Mathematics 86, 2, 1980.
[24] G. Xiang and M. Han. Global bifurcation of limit cycles in a family of multiparameter systems.
Internat. J. Bifur. Chaos 14, 3325–3335, 2004.
[25] G. Xiang and M. Han. Global bifurcation of limit cycles in a family of polynomial systems.
J. Math. Anal. Appl. 295, 633–644, 2004.
(J. Tomás Lázaro) Departament de Matemàtica Aplicada I, Universitat Politècnica

de Catalunya,, Barcelona, SPAIN
E-mail address: jose.tomas.lazaro@upc.edu
Volume 509, 2010
From the index of a differential operator to the Milnor

number of a singularity
David Mond
Abstract. In this paper I begin by recalling the outline of my Master’s thesis,

written under the direction of Jairo Charris, and go on to survey some related
ideas due to Malgrange. I end by explaining how Malgrange used the Gauss-
Manin connection to calculate the Milnor number (the rank of the vanishing
homology) of an isolated hypersurface singularity.
Contents
1. Introduction 129
2. The index of a linear differential operator 130
3. The Milnor number and the Gauss-Manin connection 134
References 141
1. Introduction
This is a meeting on Differential Equations and Orthogonal Polynomials, rather
far from my area of research, in Singularity Theory. My qualification for partic-
ipating is having had the good fortune to have been a student of Jairo Charris,
with whom I wrote my MSc dissertation in 1978. For this reason Primitivo Acosta
Humánez suggested I should at least begin by speaking about the work I did in my
dissertation. I will give a brief account of this, and then explain how it fits into a
larger landscape of singular points of differential operators. I will finish by describ-
ing some work of B. Malgrange, which appeared shortly before I began working
with Jairo, in which the notion of the index of a differential operator was used to
calculate the Milnor number of an isolated hypersurface singularity, a topic much
closer to my current interests. The talk is purely expository.
1991 Mathematics Subject Classification. Primary 54C40, 14E20; Secondary 46E25, 20C20.
Key words and phrases. Differential geometry, algebraic geometry.
The author was supported in part by Grant #000000.
2010
c 0000
c Mathematical
holder)
1
129
127
130
128
2 DAVID MOND
2. The index of a linear differential operator

Jairo Charris was my adviser when I wrote my MSc dissertation in the National
University of Colombia (Bogotá), in 1977-78. His proposal was to associate, to a
holomorphic linear differential operator
dn d
(2.1) P = an (z)n
+ · · · + a1 (z) + a0 (z),
dz dz
where the aj are holomorphic in an open set Ω of the complex plane C, an exact
sequence
P
(2.2) 0 → ker P → O(Ω) −→ O(P ) → HomZ (H1 (Ω; Z), ker P ) → 0.
I do not know where this idea came from. Presumably, as a good research director,
he knew more than he admitted. Nevertheless, the exact sequence has as special
case the theorem of Morera – if P is the operator d/dz (so ker P = C), then defining
the morphism

f ∈ O(Ω) → If ∈ HomZ (H1 (Ω; Z), C), by If ([γ]) = f (z)dz
γ
the exactness of (2.2) is equivalent to the statement that f ∈ O(Ω) is the derivative
of a holomorphic function if and only if its integral along every closed curve in Ω
is equal to 0 – Morera’s Theorem.
I was able to construct the exact sequence, but only by making use of an
additional hypothesis:
(H) that the aj should be holomorphic not only in Ω but in a simply
connected domain containing Ω, in which an should have no zero.
This guarantees that there exists a global basis ϕ1 , . . ., ϕn of solutions of the ho-
mogeneous equation
(2.3) Pg = 0
– that is, of ker P – defined in all of Ω. Without this hypothesis, ker P may even
be reduced to 0.
Supposing (H), the morphism I : O(Ω) → HomZ (H1 (Ω; Z), ker P ) can be con-
structed using the well-known method of variation of parameters: Given f ∈ O(Ω),
one looks for functions c1 , . . ., cn such that
c1 ϕ1 + ··· + cn ϕn = 0
c1 ϕ1 + ··· + cn ϕn = 0
(2.4) ··· ··· ··· ··· ···
c1 ϕ1 cn ϕn
(n−2) (n−2)
+ ··· + = 0
c1 ϕ1 cn ϕn
(n−1) (n−1)
+ ··· + = f.
It is easy to check that (2.4) implies that
(2.5) P (c1 ϕ1 + · · · + cn ϕn ) = f.
Moreover, the determinant of the matrix of coefficients of the linear system (2.4)
is the Wronskian of the n functions ϕ1 , . . ., ϕn , so it is never zero in Ω (see e.g. [1,
Section 27.5]), and the cj can be found by Cramer’s rule. The problem now is to
go from cj to cj , a non-trivial step if Ω is not simply connected, precisely because
in this case not every function is a derivative. In view of Morera’s theorem quoted
FROM INDEX OF DIFFERENTIAL OPERATOR TO MILNOR NUMBER 131
129
3
above, this suggests that for each f ∈ O(Ω) one should define the operator If in
(2.2) as the integral, along closed curves in Ω, of the functions

ϕ1 · · · ϕk−1 0 ϕk+1 · · · + ϕn

ϕ1 · · · ϕk−1 0 ϕk+1 · · · + ϕn

··· ··· ···
(n−2)
ϕ1 · · · ϕ
(n−2)
0 ϕ
(n−2)
· · · + ϕ (n − 2)
k−1 k+1 n
ϕ(n−1) · · · ϕ(n−1) f ϕ(n−1) · · · + ϕ
(n−1)
k−1 k+1 n
ck =
1
W (ϕ1 , . . ., ϕn )
which are obtained solving the system of equations(2.4):
n

If ([γ]) = ck (ξ)dξ ϕk .
k=1 γ
If If is the zero operator then each of the ck is the derivative of a holomorphic
function ck on Ω, the ck satisfy (2.5), and f is in the image of the operator P .
Slightly more explicitly, if If = 0 then
z
g(z) = ck (ξ)dξ ϕk
k z0
(the integral along any curve joining z0 to z in Ω) is independent of the choice of

curve, and is a solution of the inhomogeneous equation
P g = f.
What was hard was proving exactness at the second occurrence of O(Ω). It all
came down to proving the determinantal identity
(2.6) det(M ) det(Mji11ji22 ) = det(Mji11 ) det(Mji22 ) − det(Mji21 ) det(Mji12 )
(where M is a square matrix and the sub- and super-indices indicate omitted
columns and rows) which now, on preparing this lecture, I recognized immedi-
ately as a so-called Laplace identity (see for example [3]), which I had used for
a second time ten years later, without noticing the coincidence, in a paper about
Fitting ideals.
I spent more than a month trying to prove (2.6). After some time I even used
a desktop computer, the first in the department, to provide numerical evidence. I
was mortified when it told me the result was false, but was later reassured when
it gave different answers for the determinant of a matrix and of its transpose. I
do not know whether I gained anything from spending so long trying to prove
something that was in fact already well-known. But I have no doubt that under
such circumstances, pressing ahead blindly, even though the desired result may be
well-known, is better than giving up blindly because it may be well known.
My dissertation work on the exact sequence (2.2) was published as [11].
Around then, Jairo told me to learn something about sheaf theory. He gave me
a copy of Gunning’s Lectures on Riemann Surfaces ([5]), which begins with a careful
introduction to sheaf theory, including sheaf cohomology via the Čech complex and
as a derived functor. On reading this I realized that the exact sequence (2.2),
which had cost me so much effort to construct, was a case of a much more general
132
130
4 DAVID MOND
principle, the long exact sequence of sheaf cohomology associated to a short exact
of sheaves. If
(2.7) 0→F →G →H →0
is a short exact sequence of sheaves on a space X, then there is a long exact sequence
of cohomology
δ
(2.8) 0 → Ȟ 0 (X; F ) → Ȟ 0 (X; G ) → Ȟ 0 (X; H ) −→
δ
→ Ȟ 1 (X; F ) → Ȟ 1 (X; G ) → Ȟ 1 (X; H ) −→ · · ·
q
Here I use “Ȟ ” to denote sheaf cohomology. In my case, the short exact sequence
was just
P
0 → KP → OΩ −→ OΩ → 0
where OΩ is the sheaf of germs of holomorphic functions on Ω, and KP is the sub-
sheaf of germs of solutions to the homogeneous equation (2.3). Note that for any
sheaf F , Ȟ 0 (X, F ) is the space of its global sections, so that Ȟ 0 (Ω; OΩ ) = O(Ω)
and Ȟ 0 (Ω; KP ) = ker P . The surjectivity of P as a morphism of sheaves of germs
of holomorphic functions depends on the hypothesis that an is never zero in Ω. It
loses surjectivity in the passage from local to global – which is exactly what sheaf
cohomology is so good at dealing with.
Since, as is well known, Ȟ k (Ω; OΩ ) = 0 for k ≥ 1, the exact sequence of
cohomology (2.8) arising from the short exact sequence (2.7) is reduced to
P
0 → ker P → O(Ω) −→ O(Ω) → Ȟ 1 (Ω; K ) → 0.
By the hypothesis (H), KP is the constant sheaf ker P so that
Ȟ 1 (Ω; K ) = Ȟ 1 (Ω; ker P ) H 1 (Ω; ker P )
(the term on the right is standard cohomology, e.g. singular cohomology, with
coefficients in the C-module ker P )
HomZ (H1 (Ω; Z), ker P ),
and so from (2.8) we have obtained (2.2) – although without an explicit description
of the morphism I. But we have obtained more: the sequence
P
(2.9) 0 → ker P → O(Ω) −→ O(Ω) → Ȟ 1 (Ω; KP ) → 0
is always exact, even without hypothesis (H), so long as an has no zeros in Ω. But
how to interpret it? The (global) kernel ker P is in general equal to zero; if an has
a zero in a bounded component of C Ω, there is in general non-trivial monodromy
in the solutions of (2.3): if we use analytic continuation along a closed curve around
a zero of an to prolong a germ of solution at some point z0 , we may return with a
different solution of (2.3). In general there is no solution to (2.3) defined globally
in Ω.
The sheaf K is an example of a local system: a locally constant sheaf whose
fibre at each point is a finite dimensional vector space.
At the same time as I was working with Jairo on this question, Lucimar Nova
also was working on her master’s dissertation with him. She had taken on the deeper
question of trying to calculate the index of the differential operator (2.1) in case an
has zeros in Ω. The two of them were very disappointed to learn that B. Malgrange
had published a paper, Sur les points singuliers des equations différentielles, [7], in
which he solved the problem they were working on. Malgrange’s paper was of some
131
5
importance - the review in Mathematical Reviews of a preliminary version, [6], was

written by none other than Henri Cartan.
The linear operator L : E → F has an index if its kernel and cokernel are finite
dimensional; the index I (L), or I (L, E, F ) is then defined by
I (L) = dim ker L − dim coker L.
Given the hypothesis (H), we can calculate the index of the operator (2.1) from the
exact sequence (2.2):
I (P ) = dim ker P − dimC HomZ (H1 (Ω; Z), ker P ) =
(2.10) = dimC ker P (1 − rankH1 (Ω; Z)) = n(1 − b1 (Ω))

where b1 = rank(H1 (Ω); Z).
The index is a robust invariant. In many contexts, certain perturbations leave
the index of an operator unchanged even though they alter the dimension of its
kernel and cokernel. This makes it possible to find the index in cases where we
have no access to the dimension of its kernel or cokernel. In this, it resembles the
Euler characteristic of a topological space. Malgrange makes use of this fact when
he calculates the index in [7]. First he finds the index of the operator an (z)dn /dz n ,
an easy calculation since it is the composition of multiplication by an and the n-
fold iteration of the operator d/dz, for which it is very easy to find the kernel and
cokernel. Then he makes use of the fact that with respect to a suitable norm on the
space of functions, the missing part of the operator (2.1), namely an−1 dn−1 /dz n−1 +
· · · + a0 , is a compact operator. Adding a compact operator does not change the
index, so that the calculation for an dn /dz n gives the right answer.
This robustness of alternating sums is present also in the cohomology of local
systems. The Euler characteristic of a local system S on a space X,

χ(X, S ) := (−1)k dim Ȟ k (X, S ),
k
depends only on the rank of S , and thus is equal to

χ(X, ⊕n1 CX ) = nχ(X, CX ) = nχ(X),
where n is the rank of S and χ(X) is the topological Euler characteristic of X. So
from (2.9) it follows that
I (P ) = χ(Ω, KP ) = nχ(Ω)
– that is, (2.10) holds even without the hypothesis (H).
In [6] and [7], Malgrange incorporates a term which takes into account the
zeros of an in Ω: without any of the requirements of (H), we have
(2.11) I (P ) = n(1 − b1 (Ω)) − ν(an , Ω)
where the second term on the right is the number of zeros of an in Ω, counted with
multiplicity.
He also compared the index of the operator P on the ring C{z} of convergent
power series with its index on the ring C[[z]] of formal power series. This comparison
led to a new understanding of the crucial notion of regular singularity of a differential
operator, which, applied to the Gauss-Manin connection, plays an important role
in singularity theory.
134
132
6 DAVID MOND
3. The Milnor number and the Gauss-Manin connection

In his paper “Intégrales asymptotiques et monodromie” ([8]) Malgrange used a
related formula for the index of an operator to give an elegant proof of a formula of
Palamodov for the Milnor number of an isolated hypersurface singularity. Milnor
had proved that if f : Cn → C is a complex polynomial with isolated singularity
at z0 , and we suppose, without loss of generality, that f (z0 ) = 0, then
Theorem 3.1. (Milnor, [10]) For ε > 0 and η > 0 sufficiently small,
(1) f −1 (0) ∩ Bε (z0 ) is homomorphic to the cone on its boundary f −1 (0) ∩ Sε ,
(2) for 0 < |t| < η, Xt := f −1 (t) ∩ Bε (z0 ) (the Milnor fibre of the singularity
of f at z0 ) has the homotopy type of a wedge of spheres of dimension n−1,
(3) the map f : Bε (z0 ) ∩ f −1 (Dη∗ ) → Dη∗ is a C ∞ -locally trivial fibre bundle
(the Milnor fibration), where Dη∗ = Dη (0) {0}.
In other words,
(1) inside the ball Bε (z0 ), f −1 (0) is contractible
(2) the nearby level set f −1 (t) ∩ Bε (z0 ), has a particularly simple homotopy
type, and
(3) the union of these level sets forms a locally trivial family over Dη∗ .
Palamodov showed that the number of spheres making up the wedge (and thus the
rank of H n−1 (Xt )) is equal to
OCn ,z0
(3.1) µ := dimC ∂f ∂f
( ∂z1 , . . . , ∂zn
)
This number, which is also the number of non-degenerate critical points into which
the critical point of f at z0 splits when f is perturbed (a phenomenon known as
conservation of multiplicity), is known as the Milnor number of the singularity of
f at z0 , and is its most important numerical invariant. Several proofs of (3.1) are
known. I want now to explain the proof that Malgrange gave in [8]. The idea is to
calculate the rank of H n−1 (Xt ; C) using holomorphic de Rham cohomology.
3.1. Holomorphic de Rham cohomology. Let X = Bε (z0 )∩f −1 (Dη ). We
consider the complex
(3.2) 0 → OX → Ω1X → · · · → ΩnX → 0
of sheaves of holomorphic forms on X, where each arrow is the exterior derivative, a
C-linear but not OX -linear morphism. The complex (3.2) is acyclic (i.e. exact) by
the holomorphic Poincaré Lemma, except that d : OX → Ω1X has kernel CX (the
sheaf of germs of constant functions). In other words, (3.2) is a resolution of CX .
In fact, X is a Stein space ([4]), which means that the complex of global sections
of (3.2),
(3.3) 0 → O(X) → Ω1 (X) → · · · → Ωn (X) → 0
calculates the cohomology of X – it is not necessary to take the complex of all
smooth forms, as in the usual de Rham theorem. As X is contractible, (3.3) is
exact too, once we insert 0 → C → O(X) at the start. The Milnor fibre Xt is also
a Stein space, so its cohomology also is calculated from the complex Ω• (Xt ). Now,
ΩkX
ΩkXt = ,
(f − t)ΩkX+ df ∧ Ωk−1
X
133
7
and the ΩkXt form a family over D := Dη ,

ΩkX
ΩkX/D = ,
df ∧ Ωk−1
X
such that
ΩkX/D
ΩkXt =
mD,t ΩkX/D
where mD,t is the maximal ideal in OD,t . The ΩkX/D form the relative holomorphic
de Rham complex,
(3.4) Ω•X/D := 0 → OX → Ω1X/D → · · · → ΩnX/D → 0,
where the arrow ΩkX/D → Ωk+1 X/D is just the exterior derivative, which passes to the
quotient; it is usually denoted by dX/D . Note that dX/D (ω) = 0 if and only if there
exists σ such that
dω = df ∧ σ.
The relative differential dX/D is not OX -lineal, but it is linear over OD via f :
denoting the coordinate in D by t, we have
d(t · ω) = d(f ω) = f dω + df ∧ ω = f dω = t · dω mod df ∧ ΩkX .
This OD -lineality means that the cohomology of the complex Ω•X/D is an OD -
module.
We need one further step in this construction, the push-forward to D, via f , of
the complex Ω•X/D . For each q ∈ N we define the push-forward sheaf f∗ (ΩqX/D ) by
taking, as sections over U ⊂ D, the sections of ΩqX/D over f −1 (U ). In an obvious
way the exterior derivative makes the collection of push-forwards f∗ (Ω•X/D ) into a
complex. If t ∈ D is a regular value of f (i.e. if t = 0) then the stalk of f∗ (ΩqX/D )

at t is isomorphic to Ωq (Xt ) ⊗C OD,t , and moreover Hq f∗ (Ω•X/D,t ) , the q-th

cohomology of the complex f∗ (Ω•X/D )t , is isomorphic to H q (Ω• (Xt ) ⊗C OD,t .
For t = 0, Xt is non-singular and a Stein space, so that
(3.5) H k (Ω• (Xt )) H k (Xt ; C).
The term on the left of (3.5) can be recovered from the cohomology of the pushed-
forward complex f∗ (Ω•X/D ):
Ωq (Xt ) = f∗ (ΩqX/D )t /mD,t · f∗ (ΩqX/D )t
and so
Hk (f∗ (Ω•X/D )t )
(3.6) H k (Ω• (Xt ))) = .
mD,t · Hk (f∗ (Ω•X/D )t )
The sheaf of OD -modules Hk (f∗ (Ω•X/D )) is coherent – a famous theorem of Brieskorn

[2]. Its restriction to D {0} is a sheaf of free OD -modules; by (3.5) and (3.6), we
have
(3.7) rank Hk (f∗ (Ω•X/D )t ) = dimC H k (Xt ; C).
We will now use an index theorem to calculate its rank.
136
134
8 DAVID MOND
Recall that if E and F are OC,0 -modules, a C-linear map D : E → F is a

connection if for ω ∈ E and g ∈ OC,0 ,
dg
(3.8) D(gω) = ω + gDω.
dt
The index theorem we will use is the following:
Theorem 3.2. (Malgrange [8, 2.3]) Let D : E → F be a connection, where E
and F are OC,0 -modules of finite type, with E ⊂ F and F/E a torsion module (i.e.
for every m ∈ F there is a k ∈ N such that tk m ∈ E). Then
(1) The kernel and cokernel of D are of finite dimension, and
(2)
I (D) + dimC F/E = rank E.
The theorem follows easily from (2.11).
As a trivial example: D = d/dt : OC,0 → OC,0 is surjective and has kernel
of dimension 1, so that its index is 1. In this case E = F and is of rank 1, so the
equality of (3.2(2)) does indeed hold.
3.2. Heuristics of the Gauss-Manin connection. We now look briefly

at the orgins of the connection used by Malgrange when he applied 3.2 to prove
Palamodov’s formula. This heuristic discussion is not needed for the proof, but
may help to understand it.
Let p : X → B be a locally trivial fibre bundle with connected base B. For
U ⊂ B it is convenient to denote by XU the set p−1 (U ). For each t ∈ B and each
fixed q ∈ N we consider the cohomology space H q (Xt ; C). Since B is connected, all
of the fibres Xt are diffeomorphic to one another, and therefore for each fixed q all
of the cohomology spaces H q (Xt ; C) are isomorphic to one another. Their disjoint
union

H q (Xt ; C)
t∈B
can be given the structure of a complex vector bundle over B, H q (X/B), in a

natural way. Its local trivialisations arise as follows: if U is a contractible open set
in B, then for every t ∈ U , Xt is a deformation retract of XU . It follows that the
morphism of cohomology
i∗t,U : H q (XU ) → H q (Xt )
induced by the inclusion Xt → XU is an isomorphism. This family of isomorphisms
defines a trivialisation of H q (X/B)|U = H q (XU /U ). It is easy to check that the
crossover maps between any two such local trivialisations of H q (X/B) are locally
constant. The local trivialisations thus give rise to a flat connection on the sheaf
of sections of the bundle, the Gauss-Manin connection, which is defined as follows.
To determine a connection it is enough to say, for each t ∈ B, which sections are
infinitesimally horizontal at t – in other word, which are the “horizontal” subspaces
of the tangent spaces to the total space at each point. This data is known as an
Ehresmann connection. In the case of the Gauss-Manin connection, we can do
more than that. We can exhibit, for each t ∈ B, a neighborhood U of t and an
O(U )-basis for the space of holomorphic sections of H q (X/B) over U , consisting
135
9
of sections which are horizontal at every point. Given U ⊂ B, every cohomology

class [ω] ∈ H q (XU ) defines a section s[ω] of H q (X/B) over U ,
s[ω] (t) = i∗t,U ([ω]),
and we take these as the horizontal sections. If U is contractible, the rank of
H q (XU ) is equal to the rank of H q (Xt ) for each t ∈ U , so these sections do indeed
generate the space of all holomorphic sections of H q (X/B) over OU .
We now apply this construction to the locally trivial fibre bundle f : X
X0 → D {0} of Milnor’s theorem 3.1. It leads us to a connection on the cohomol-
ogy of f∗ (Ω•X ∗ /D∗ ), where for brevity we write D∗ = D {0} and X ∗ = X X0 .
Crucially, it leads also to a “singular” connection on the cohomology of f∗ (Ω•X/D )
— we do not need to exclude the singular point. That the connection becomes
singular means in this case that the covariant derivative of a germ of holomorphic
section at 0 may not itself be holomorphic. In fact it will be meromorphic; the
nature of its pole at 0 is important in other contexts, and is described by the term
“regular singularity”, about which I say no more here.
A germ of section of Hq (f∗ (Ω•X/D )) at t ∈ D is given by a holomorphic q-form
ω defined in some neighborhood of Xt in X, which is closed in f∗ (ΩqX/D )t ; that is,
such that
(3.9) dω = df ∧ σ

for some q-form σ. This means that d ω|Xt ) = 0 for each t, so that ω|Xt defines a
cohomology class [ω|Xt ] in H q (Xt : C), and the assignment

sω : t → ω|Xt
defines a germ of holomorphic section of the cohomology bundle H q (X ∗ /D∗ ) at t.
By the reasoning of the previous paragraph, this section is horizontal if and only
if it comes from a cohomology class on some neighborhood of Xt in X. And this
is the case if and only if ω is not only closed in the relative complex Ω•X/D but in
the absolute complex Ω•X . That is, sω is horizontal if and only if dω = 0, that is,
if and only if (3.9) holds with σ = 0. For this reason, the natural definition of the
covariant derivative of sω as section of H q (X ∗ /D∗ ) is the section defined by sσ .
By the natural identification of Hq (f∗ (Ω•X/D ))t with the space of germs at t of
holomorphic sections of H q (X ∗ /D∗ ), this leads us to define the covariant derivative
in Hq (f∗ (Ω•X/D ))t by
(3.10) ∇d ω = σ.
dt
It is necessary to appreciate that σ is not uniquely defined –

(3.11) dω = df ∧ σ =⇒ ∀ρ ∈ Ωq−1
X , dω = df ∧ (σ + df ∧ ρ).
This means that we could just as well choose σ + df ∧ ρ in place of σ. But in the
space of relative forms, σ and σ + df ∧ ρ are equal, so that this imprecision in the
choice of σ does not matter.
All of this is valid in the complex of germs f∗ (Ω•X/D )t for t = 0. As we have
seen, it relies on
(1) the fact that Xt is a smooth Stein space, so that the “holomorphic de
Rham theorem” (3.5) holds;
(2) the fact that over some neighborhood U of t, p : X → D is diffeomorphic
to a projection U × Xt → U .
138
136
10 DAVID MOND
Slightly surprisingly, the Gauss Manin connection can also be defined, by (3.10), on
the cohomology of the complex f∗ (Ω•X/D )t for t = 0, at the cost of acquiring a pole.
It is here that it is used by Malgrange. More precisely, contractibility of X and
of X0 implies that this cohomolology reduces to that of the (un-pushed-forward)
complex of germs of holomorphic forms Ω•X/D,z0 , and it is this complex that we
now look at in detail.
First, the fact that f has isolated singularity implies that the imprecision (3.11)
in the choice of σ is the only imprecision. For if dω = df ∧ σ1 = df ∧ σ2 then
(3.12) df ∧ (σ1 − σ2 ) = 0,
and this implies that
(3.13) σ1 − σ2 = df ∧ ρ
for some q − 1-form ρ, as in (3.11). The fact that (3.12) implies (3.13), for 1 ≤ q ≤
n−1, is a consequence of acyclity of the Koszul complex on the partial derivatives of
•
f . For the complex (Ω•X,x , df ∧) is isomorphic to the Koszul complex ( OnX,x , df ∧),
0 1 n−1 n
df df df df
(3.14) 0→ ΩnX,x −→ ΩnX,x −→ · · · −→ OnX,x −→ OnX,x
and provided f either is non-singular at x, or has isolated singularity at x, the
Koszul complex is well-known to be acyclic (see e.g. [9, Chapter 6]). So the
operator ∇d/dt is well defined even for t = 0. I leave the reader to check that ∇d/dt
is a connection, in the sense that (3.8) holds.
What is not immediately clear is that ∇d/dt ω, the form σ such that dω = df ∧σ,
should be closed in the relative complex (3.4). If q < n − 1, or if t = 0, this is easy
to see: suppose that ω is a relatively closed q-form. Then from
dω = df ∧ σ
we get
(3.15) 0 = ddω = −df ∧ dσ.
Since dσ is a q + 1 form and q + 1 < n, the acyclicity of the Koszul complex implies
that dσ itself is equal to df ∧ ρ for some ρ, so that dX/D σ = 0 as required. If
q = n − 1 and x = 0, then dσ is necessarily equal to df ∧ ρ for some n − 1-form
ρ, simply because the last morphism in the Koszul complex is surjective at points
where f is non-singular.
However, when q = n − 1 and x = z0 , dX/D σ is not necessarily equal to zero.
Equation (3.15) still holds, but is now an empty statement because dσ is already
an n-form, and the wedge of df with any n-form is zero. We cannot conclude from
(3.15) that dσ = df ∧ ρ for some n − 1- form ρ. In fact the non-surjectivity of
df ∧ : Ωn−1 n
X,0 → ΩX,0 is crucial for the application of Malgrange’s index theorem, as
we will see in the next section.
3.3. Calculation of the Milnor number. By the contractibility of the cen-

tral fibre, 3.1(1),
Hq (f∗ (Ω•X/D )0 ) = Hq (Ω•X/D,z0 )
and as the rank of a coherent sheaf is locally constant, it is enough, for our purposes,
to calculate the rank of Hq (Ω•X/D,z0 ). We distinguish two cases:
137
11
Case 1: q ≤ n − 2: In this case, the σ which we choose in (3.10) is itself closed

in the relative complex. So ∇d/dt defines a connection
δq : Hq (Ω•X/D,z0 ) → Hq (Ω•X/D,z0 ).
From this we deduce that Hq (Ω•X/D,z0 ) is a free OD,0 - module. For given any
relation (with coefficients in OD,0 ) among a set of generators, applying ∇d/dt we
obtain a relation of lower order, and thus, eventually, a relation in which there is a
coefficient which is a unit in OD,0 . Such a relation cannot exist among a minimal
set of generators, and so among a minimal set of generators there can exist no
relation at all. As δq is a non-singular first order differential operator on a free
OD,0 -module, it is surjective. An argument in the next section shows that it is
also injective, except in case q = 0, when it has kernel equal to the set of constant
functions C. It follows that its index is 1 for q = 0 and 0 for 1 ≤ q ≤ n − 2. We
conclude, by 3.2, that
1 if q = 0
rank Hq (Ω•X/D,z0 ) = .
0 if 1 ≤ q ≤ n − 2
and therefore that

1 if q = 0
dimC H q (Xt ; C) =
0 if 1 ≤ q ≤ n − 2
Case 2: q = n − 1: In this case σ does not necessarily belong to Z n−1 (Ω•X/D,z0 ).

The most we can say is that σ ∈ Ωn−1
X/D,z0 . We define
Z n−1 (Ω•X/D,z0 ) Ωn−1

X/D,z0
E= = Hn−1 (Ω•X/D,z0 ), F = .
dΩn−2
X/D,z0 dΩn−2
X/D,z0
The Gauss-Manin connection gives rise to a connection δn−1 : E → F . By an

argument explained in the next section, δn−1 is an isomorphism, so I (δn−1 ) = 0.
On the other hand, 3.2 says that
F
I (δn−1 ) + dimC = rank E.
E
The exterior derivative gives rise to an isomorphism
F Ωn−1
X/D,z
= n−1 • 0 ΩnX/D,z0
E Z (ΩX/D,z0 )
and finally,
ΩnX,z0 OX,z0
ΩnX/D,z0 = .
df ∧ Ωn−1
X,z0
(∂f /∂z1 , · · ·, ∂f /∂zn )
As I (δ) = 0, this shows that
OX,z0
rank E = dimC ,
(∂f /∂z1 , · · ·, ∂f /∂zn )
and we conclude by (3.7) that this is the dimension of H n−1 (Xt ; C).
140
138
12 DAVID MOND
3.4. The Gauss-Manin connection as connecting homomorphism in

a long exact sequence. The fact that f has isolated singularity at 0 means that
for every q-form σ ∈ ΩqX,z0 with q < n,
(3.16) df ∧ σ = 0 ⇔ σ ∈ df ∧ Ωq−1
X .
In other words,
df ∧
ΩqX/D,z0 −→ df ∧ ΩqX,z0
is an isomorphism for q < n. This isomorphism commutes with the exterior deriv-
ative (dX/D on the left, d on the right), so there is an isomorphism for all q
(3.17) Hq (Ω̄•X/D,z0 ) Hq+1 (df ∧ Ω•X,z0 )
where the complex Ω̄•X/D is just Ω•X/D except that its last term ΩnX/D is replaced
by 0.
Now consider the short exact sequence of complexes
0 → df ∧ Ω•X,z0 → Ω•X,z0 → Ω•X/D,z0 → 0.
It gives rise to a long exact sequence of cohomology. As Ω•X,z0 is exact except at

Ω0 , where it has cohomology equal to C, this long exact sequence breaks up into a
short exact sequence
δ̃
(3.18) 0 → C = H0 (Ω•X,z0 ) → H0 (Ω•X/D,z0 ) −→
0
H1 (df ∧ Ω•X,z0 ) → 0
and a collection of isomorphisms

δ̃
(3.19) Hp (Ω•X/D,z0 ) −→
n
Hp+1 (df ∧ Ω•X,z0 ).
In each case δ̃q is the connecting homomorphism of the long exact sequence. Let δ̄q
denote the composite of δ̃q with the isomorphism of (3.17). From (3.18) and (3.19)
we obtain, respectively, an exact sequence
δ̄
(3.20) 0 → C = H0 (OX,z0 ) → H0 (Ω•X/D,z0 ) −→
0
H0 (Ω̄•X/D,z0 ) → 0
and isomorphisms
δ̄q
(3.21) Hq (Ω•X/D,z0 ) −→ Hq (Ω̄•X/D,z0 ).
for 1 ≤ q ≤ n − 1. Note that

Ωn−1
X/D,z0
Hn−1 (Ω̄•X/D,z0 ) =
dΩn−2
X/D,z0
is the module F of the previous section, so the isomorphism δ̄n−1 of (3.21) has the
same source and target as the singular Gauss Manin connection δn−1 : E → F of
the previous section. In the last section we asserted that δq is an isomorphism for
1 ≤ q ≤ n. The proof that this is so is simply the fact that δ̄q = δq . This can be
easily checked by following the definition of connecting homomorphism. I urge the
reader to do this!
139
13
References
[1] V.I. Arnold, Ordinary Differential Equations, MIT Press, 1973
[2] E. Brieskorn, Die Monodromie der isolierten Singularitäten von Hyperflächen, Manuscripta
Math. 2, (1970) 103-161
[3] C. de Concini, D. Eisenbud, C. Procesi, Young diagrams and determinantal varieties, Invent.
Math. 56 (1980), 129-165
[4] H. Grauert and R. Remmert, Theory of Stein Spaces, Grundlehren der mathematischen Wis-
senschaften 236, Springer-Verlag, 1979
[5] R.C. Gunning, Lectures on Riemann Surfaces, Princeton Mathematical Notes, Princeton
University Press, 1966
[6] B. Malgrange, Remarques sur les points singuliers des équations différentielles, Comptes Ren-
dus de l’Acad. Sci. de Paris Sér. A-B 273 (1971) A1136-A1137
[7] B. Malgrange, Sur les points singuliers des équations différentielles, L’Enseignement
Mathématique 20 (1974) 147-176
[8] B. Malgrange, Intégrales asymptotiques et monodromie, Ann. Ecole Normale Superieure (4)
1974, 405-430
[9] H. Matsumura, Commutative ring theory, Cambridge Studies in Advanced Mathematics 8,
Cambridge University Press, 1986
[10] J. Milnor, Singular points on complex hypersurfaces, Princeton University Press, 1968
[11] D. Mond, Sobre la cohomologı́a asociada a un operador lineal diferencial complejo, Revista
Colombiana de Matemáticas, XIII, 3, 1979, 171-192.
[12] D. Mond, Differential forms on free and almost free divisors, Proc. London Math. Soc. (3)
81 (2000), 587-617
[13] D. Mond, Aspectos topológicos de la deformaciones de singularidades, Revista Colombiana
de Matemáticas, 2007.
Mathematics Institute, University of Warwick, Coventry CV4 7AL, England

E-mail address: d.m.q.mond@warwick.ac.uk
Volume 509, 2010
Integrability of Dynamical Systems through Differential

Galois theory: a practical guide
Juan J. Morales-Ruiz and Jean-Pierre Ramis
... Quan arribis, si et sento, hauré de saludar-te amb un gran crit.

Car moro sense cap saviesa, però molt ric de passos de perdut vianant.
(Salvador Espriu, Thànatos)
In memoriam of our friend Josep-Marı́a Peris (1951 – 2006)
Abstract. We survey recent advances in the non-integrability criteria for

Hamiltonian Systems which involve the differential Galois group of variational
equations along particular solutions. The emphasis is on algorithms and ap-
plications, not theory. For most applications one does not need a deep un-
derstanding of the differential Galois theory. All essentials are presented here,
along with numerous concrete examples.
Contents
1. Introduction 144
2. General Non-integrability Theorems 146
3. Homogeneous Potentials and Related Problems 157
4. Hamiltonian Rigid Body Problem 189
5. Cosmological Models 192
6. An Application to Painlevé’s Transcendents 195
Appendix A. Algorithmic Considerations 201
Appendix B. Hypergeometric Equation 211
Appendix C. Lamé Equation 212
References 215
1991 Mathematics Subject Classification. Primary 70H06, 70H07, 70H33, 70F07, 70F10; Sec-
ondary 34A05, 34A30, 34C14, 34M15, 34M35.
Key words and phrases. Hamiltonian Systems, Integrability, Variational Equations, Differ-
ential Galois Theory.
The research of the first author has been partially supported by grant MCyT-FEDER
MTM2006-00478 of Spanish government.
1
143
141
144
142
2 JUAN J. MORALES-RUIZ AND JEAN-PIERRE RAMIS
1. Introduction
It has been several years since our papers [98, 97, 99] and the book [95] were
published; in the meantime, new lines of research have been opened, new results
have been obtained by several authors, and old results are included in a natural
way in the framework. Some of them are the following:
a) A proof of the conjecture about the higher order variational equations.
b) The obtention of new results of a global nature as oriented to the classi-
fication of integrable cases of homogeneous polynomial potentials.
c) New non-integrability results for several N -body problems.
d) New non-integrability results for several cosmological models.
e) Non-integrability results for other specific families of systems such as
Painlevé’s transcendents, including new simple proofs of old results, e.g.
for the rigid body.
f) Obstructions to the existence of real analytical first integrals.
g) New contributions about connections of our approach with chaotic dy-
namics – more specifically, splitting of separatrices.
h) The proposal of some extensions to non-holonomic mechanical systems,
control theory, and other not necessarily Hamiltonian systems.
Our present aim is to survey these new results. Due to limitations of space it
was impossible to give a complete account of all the above items. The choice of
the topics was subjective, a fact for which we apologize. In particular, topics f),
g) and h) will be not mentioned within the text. Readers interested in f) can read
the papers by Ziglin ([150]) and Audin ([11]). Topic g) started over eight years
ago in a joint publication by Josep-Maria Peris and one of the authors ([96]) and
was followed by Yagasaki and others ([143, 144, 145]). Topic h) is now being
a very active research field under Tsygvintsev, Dullin, Maciejewski, Przybylska,
Respondek, Weil, and others; as some of the fundamentals of this area are still not
completely finished, although some important results are already obtained, we only
give a few references. In [32], the obstructions to the existence of analytical first
integrals to the Rattleback problem, a difficult non-Hamiltonian and non-holonomic
rigid mechanical problem, are studied. Reference [71] features the study of the non-
integrability of a non-holonomic Hamiltonian problem: the Suslov problem. Paper
[81] is devoted to proving the non-integrability of a sub-Riemannian problem which
is important in control theory.
The emphasis will be put on the applications and our approach will essentially
be about methodology and algorithms. We also made a considerable effort to remain
at a relatively elementary level and to write a self-contained text. This contribution
is also the answer to some colleagues concerning the need of an introductory text in
our field. Thus, for an important part of the text we assume virtually no differential
Galois prerequisite from the reader. Whoever is interested can find some of the main
results of this theoretical framework in the book [95] or an elementary introduction
with proofs in the monograph of Audin [9], or else in our original articles [97, 98,
99, 102].
As the reader may check, applications are possible through a unified and sys-
tematic approach:
(1) Selecting a particular integral curve.
(2) Computing the VE.
DYNAMICAL SYSTEMS AND DIFFERENTIAL GALOIS THEORY 145
143
3
(3) Checking whether the identity component of the differential Galois group
of the VE is commutative.
Step 2 is easy, as we will see. Step 3 is generally quite involved. Fortunately, for
particular cases occurring in many applications, some efficient algebraic algorithms
do exist. The prototype is Kovacic’s algorithm for second-order equations. In nearly
all applications known to the authors, Step 1, common to all classical proofs of non-
integrability, is achieved due to the existence of a completely integrable subsystem,
typically due to the presence of an invariant plane.
In a joint work of the authors with Simó, the above method has been general-
ized to the higher order variational equations, V Ek , where their solutions are the
quadratic, cubic, ... contributions to the Taylor series of the flow along the partic-
ular integral curve. Hence, in the above Steps 2 and 3 we can replace V E1 := V E
by V Ek .
Finally, and although the numerous results and methods in this contribution
are nowadays collected under the umbrella name “Morales-Ramis theory”, it is the
authors’ contention that a more proper denomination should be “Ziglin-Morales-
Ramis theory”, since it was Ziglin who first introduced the monodromy group
approach to the variational equations as a fundamental tool for obtaining, and
masterfully applying, necessary conditions to Hamiltonian integrability. Thus, in
1982 Ziglin stated his fundamental theorem about the monodromy group of the
variational equations for Hamiltonian systems in presence of meromorphic first
integrals.
Theorem 1.1 ([148]). Assume the Hamiltonian system XH admits n − m

additional analytical first integrals, independent over a neighborhood of Γ (but not
necessarily on Γ itself ). Assume moreover that the monodromy group of the normal
variational equation ( NVE) contains a non-resonant transformation g. Then any
other element of the monodromy group of the NVE sends eigendirections of g into
eigendirections of g.
(see Section 2.3 for the necessary definitions and notations)

Bearing in mind that the monodromy group is contained in the symplectic
group, we recall that a linear transformation g ∈ Sp(2n, C) is resonant if there
exist integers r1 , ..., rn , not all of which equal zero, such that λr11 · · · λrnn = 1, where
we denoted the eigenvalues of g as λi , λ−1 i .
This said, in 1983 Ziglin applied Theorem 1.1 to the non-integrability of several
Hamiltonian systems, in particular solving the integrability problem of the rigid
body with a fixed point completely ([149]).
The connection of Theorem 1.1 with our theorems in Section 2.3 comes from
the fact that, given a linear differential system, the Galois group contains the mon-
odromy group, and is actually its Zariski closure if the system is Fuchsian – see
Section 2.2. We also remark that for n − m ≥ 2 Ziglin’s in Theorem 1.1 does not
assume the complete integrability of the Hamiltonian system, since the latter trait
assumes the involutivity of the first integrals – see Section 2.3.
We would like to thank Sergi Simon for remarks concerning Section 3.4 as well
as for linguistic help. Special thanks are also due to David Blázquez for remarks
about Section 3.3. Also thanks to the anonymous referee for their constructive
remarks.
146
144
2. General Non-integrability Theorems

2.1. Algebraic groups. The minimum necessary results of linear algebraic
groups are presented. An introduction to linear algebraic groups is given in [22].
For more information see the monographs [47, 21].
A linear algebraic group G (over C) is a subgroup of GL(m, C) whose matrix
coefficients satisfy polynomial equations over C. It has both the structure of a non-
singular algebraic variety and that of a group, both structures being compatible in
that the group operation and inversion are morphisms of algebraic varieties. We
note that in a linear algebraic group there are two different topologies: the Zariski
topology, where the closed sets are the algebraic sets, and the usual Hausdorff
topology. In particular, an algebraic group is a complex analytical Lie group and
we can consider its Lie algebra. Therefore the dimension of G is the dimension of the
Lie algebra of G. Given a linear algebraic group, the maximal connected subgroup
G0 which contains the identity is an algebraic group called the identity component
of G. For those familiar to Lie theory, and since G is a complex analytical Lie
group due to satisfying algebraic equations, the underlying group of G0 coincides
with the identity component of G considered as a complex analytical Lie group. We
remark that an algebraic linear (or affine) group G is usually defined as an affine
algebraic variety with a group structure, with the above compatibility condition as
to group multiplication and taking of inverses. Then, there is a rational faithful
representation of G as a closed subgroup of GL(m, C), for some m, and we obtain
the equivalence with our definition.
It is clear that the classical linear complex groups are linear algebraic groups.
For instance SL(n, C), SO(n, C) (rotation group) and Sp(n, C) ⊂ Gl(2n, C) (sym-
plectic group) are linear algebraic groups, since they are defined by polynomial
identities.
Proposition 2.1. The identity component G0 of a linear algebraic group G is
a closed (with respect to the above two topologies) normal subgroup of G of finite
index and is connected with respect to the above two topologies. Furthermore G/G0
is a finite group given by the classes of the irreducible connected components of G.
We note that by the above proposition G0 is also a linear algebraic group and
the Lie algebra of G, Lie(G) = G coincides with the Lie algebra of G0 , Lie(G0 ) = G.
As for every Lie group, G0 is solvable or commutative if, and only if, G is solvable,
respectively commutative. Furthermore, G is connected if, and only if, G = G0 .
The characterization of the connected solvable linear algebraic groups is given
by the Lie-Kolchin theorem.
Theorem 2.2 (Lie-Kolchin Theorem). A connected linear algebraic group is
solvable if, and only if, it is conjugate to a triangular group.
Given a subset S ⊂ GL(n, C), let M be the group generated by S and let G be
the Zariski closure of the group M . By definition the group G is a linear algebraic
group; we say G is topologically generated by M . All through the text, M will
be typically the monodromy group of a Fuchsian linear differential equation and G
will be its Galois group.
Since most of the examples of irreducible equations appearing later on are
second-order and symplectic, we end this section with a classification of the alge-
braic subgroups of SL(2, C) = Sp(1, C). We shall need two lemmas.
145
5
Lemma 2.3 ([55]). Let G be an algebraic group contained in SL(2, C). Assume
that the identity component G0 of G is solvable. Then G is conjugate to one of the
following types:
(1) G is finite,

λ 0 0 −β −1 ∗
(2) G = , λ, β ∈ C ,
0 λ−1 β 0
(3) G is triangular.
Lemma 2.4. Let G be an algebraic subgroup of SL(2, C) such that the identity
component G0 is not solvable. Then G = SL(2, C).
The last lemma is well–known and follows easily from consideration of the Lie
algebra of G ⊂ SL(2, C). Indeed, if G0 is not solvable then the dimension of G
must be equal to 3, because all 2-dimensional Lie algebras are solvable.
Proposition 2.5 ([93]). Any algebraic subgroup G of SL(2, C) is conjugate to
one of the following types:

1 0
(1) Finite, G0 = {1}, where 1 = .
0 1

1 0
(2) G = G = 0
,µ∈C .
µ 1

λ 0
(3) Gk = , λ is a k-root of unity, µ ∈ C ,
µ λ−1

1 0
G0 = ,µ∈C .
µ 1

λ 0
(4) G = G0 = , λ ∈ C∗ .
0 λ−1

λ 0 0 −β −1 ∗
(5) G = , λ, β ∈ C ,
0 λ−1 β 0

λ 0 ∗
0
G = ,λ∈C .
0 λ−1

λ 0 ∗
(6) G = G0 = , λ ∈ C , µ ∈ C .
µ λ−1
(7) G = G0 = SL(2, C).
The above proposition is analogous to that found in [59], page 7. However,
we not only need to know when the identity component of the Galois group is
solvable, but when it is commutative. We remark that the identity component G0
is commutative in cases (1)–(5) and solvable in cases (1)–(6).
2.2. Picard-Vessiot Theory. At the end of the nineteenth century, Picard

([111, 112], [113, Chapitre XVII]) and, in a clearer way, Vessiot in his PhD Thesis
([140]), created and developed a Galois theory for linear differential equations.
This field of study, henceforth called Picard-Vessiot theory, was continued from the
forties to the sixties of the twentieth-century by Kolchin , through the introduction
of the modern algebraic abstract terminology and the obtention of new important
148
146
results, see [58] and references therein. Today the standard reference of this theory
is the monograph [139].
In the last years, a new revival of interest in the differential Galois theory
is being observed. This is partially due to the connections and applications to
other areas of mathematics: number theory [15, 56], asymptotic theory [86], non-
integrability of dynamical systems, etc. Here we are interested in the applications
to non-integrability. As we shall see, within differential Galois theory there is a
very nice concept of “integrability”, i.e., solutions in closed form. Furthermore,
all information about the integrability of the equation is coded in the identity
component of the Galois group: a linear equation is integrable if, and only if, the
identity component of its Galois group is solvable. We only review the necessary
definitions and results of the Picard-Vessiot theory in order to understand the
applications to non-integrability. For more information see [139].
A differential field K is a field with a derivative (or derivation) δ = , i.e., an
additive mapping satisfying the Leibniz rule. The only case we are interested in
is K = M(Γ), the meromorphic functions over a connected Riemann surface Γ.
The reason for this notation will be clear below: Γ − Γ will be the set of singular
d
points of the linear differential equation, i.e., poles of the coefficients with dt as
derivation, t being a local coordinate over the Riemann surface Γ. A particular
classical case is when K = C(t) = M(P1 ) is the field of rational functions, i.e., the
field of meromorphic functions over the Riemann sphere P1 . Another interesting
example for the applications is a field of elliptic functions.
We can define differential subfields and differential extensions in a direct way by
requiring that inclusions commute with the derivation. Analogously, a differential
automorphism in K is an automorphism commuting with the derivative. The field
of constants of K is the kernel of the derivative. In the above examples C is such
a kernel. From now on we will assume this is the case.
Let
(2.1) ξ = Aξ, A ∈ M at(m, K)
be a linear differential equation. We now proceed to associate to (2.1) the so-

called Picard-Vessiot extension of K. The Picard-Vessiot extension L of (2.1) is
an extension of K, such that if u1 , ..., um is a “fundamental” system of solutions
of the equation (2.1) (i.e., linearly independent over C), then L = K(uij ) (rational
functions in K in the coefficients of the “fundamental” matrix U = (u1 · · · um )
). This is the extension of K generated by K together with uij . We observe that
L is a differential field (by (2.1)). The existence and unicity of the Picard-Vessiot
extensions was proven by Kolchin. In the analytical case K = M(Γ), = d/dt,
this result is essentially the existence and uniqueness theorem for linear differential
equations and it is already in Vessiot [140].
As in classical Galois theory of algebraic equations, we define the Galois group
of (2.1), G := GalK (L) = Gal(L/K), as the group of all the (differential) auto-
morphisms of L leaving the elements of K fixed. Then one of the main results of
the theory is that the Galois group of (2.1) is faithfully represented as an algebraic
linear group over C, the representation is given by the action σ ∈ G,
σ(U ) = U Bσ ,
147
7
Bσ ∈ GL(m, C). The other fundamental result is the Galois correspondence be-
tween algebraic subgroups and intermediate extensions.
Theorem 2.6. Let L/K be the Picard-Vessiot extension associated to a linear
differential equation. Given any subgroup H ⊂ G := GalK (L), let KH denote the
subfield of L consisting of those elements fixed by H. Then the mapping H → M :=
KH restricts to a bijection between between the algebraic subgroups of G and the
intermediary differential fields K ⊂ M ⊂ L. Furthermore, we have
(i) To the algebraic subgroups H ⊂ G := GalK (L) correspond the Picard-
Vessiot extensions L/KH .
(ii) The group H is a normal algebraic subgroup of G if, and only if, the ex-
tension KH /K is a Picard-Vessiot extension. Then the group G/H is a linear
algebraic group and G/H = GalK (KH ).
(iii) For an arbitrary subgroup H ⊂ G the group GalKH (L) is the Zariski closure
(over the complex field C) of H.
As a corollary, when we consider the (relative) algebraic closure K of K in L,
we obtain GalK (K) = G/G0 , where G0 = GalK (L) is the identity component of
the Galois group G corresponding to the transcendental part of the Picard-Vessiot
extension, i.e., by definition, the extension L/K is the maximal transcendental
extension among those L/L1 , L1 being an extension of K. If K = K (i.e., if
G = G0 ), we say L/K is a purely transcendental extension.
We call a linear differential equation integrable if we can obtain its Picard-
Vessiot extension K ⊂ L and, hence, its general solution, by adjunction to K of
integrals, exponentials of integrals or algebraic functions of elements of K. In other
words, there exists a chain of differential extensions K1 := K ⊂ K2 ⊂ · · · ⊂ Kr :=
L, where each extension is given by the adjunction of one element a, Ki ⊂ Ki+1 =
Ki (a, a , a , ...), such that a satisfies one of the following conditions:
(i) a ∈ Ki ,
(ii) a = ba, b ∈ Ki ,
(iii) a is algebraic over Ki .
Then, it can be proven that a linear differential equation is integrable if, and only
if, the identity component G0 of the Galois group is a solvable group. In particular,
if G0 is commutative, the equation is integrable.
The usual terminology for integrable linear equations is that the associated
Picard-Vessiot extension is a Liouville extension [55]. We prefer to use a terminol-
ogy in agreement with our dynamical approach and with the creators of the theory
([140]).
Furthermore, by a classical theorem credited to Schlesinger, the relation be-
tween the monodromy and the Galois group is as follows. Let Γ − Γ be the set of
singular points of the equation i.e., the poles of the coefficients on Γ. We recall that
the monodromy group of the equation is the subgroup of the linear group defined
as the image of a representation of the fundamental group π1 (Γ) into the linear
group GL(m, C). This representation is obtained by analytical continuation of the
solutions along the elements of π1 (Γ). The monodromy group M is contained in the
Galois group G and if the equation is Fuchsian (i.e., it has regular singular points
only), then M is Zariski dense in G, see for instance [139]. In particular, this implies
that for Fuchsian differential equations the Galois group is solvable or commutative,
if, an only if, the monodromy group is solvable or commutative, respectively. In
150
148
the general case, the second author found a generalization of the above and, for
example, he showed that the Stokes matrices associated to an irregular singularity
belong to the Galois group, see [86].
We would like to point out that in the last few years a new non-linear differential
Galois theory has come into being ([83, 84, 23, 138]. The authors are convinced
that this theory will play an important role in the context of the integrability of
dynamical systems.
2.3. Non-integrability by First Order Variational Equation . Given a

dynamical system,
(2.2) ż = X(z),
with a particular integral curve z = φ(t), at the end of the nineteenth century
Poincaré introduced the variational equation (VE ) along z = φ(t),
(2.3) ξ̇ = X (φ(t))ξ,
as the fundamental tool to study the behavior of (2.2) in a neighborhood φ(t) [109].
Equation (2.3) describes the linear part of the flow of (2.2) along z = φ(t).
We have the following General Principle:
General Principle: If we assume that the dynamical system (2.2) is “integrable”
in any reasonable sense, then it is natural to conjecture that the linearized differen-
tial equation (2.3) must be also “integrable”.
It seems clear that in order to convert this principle in a true conjecture it is

necessary to clarify what kind of “integrability ” is considered for equations (2.2)
and (2.3).
As (2.3) is a linear differential equation, it is natural to consider the integrability

of this equation in the context of the Galois theory of linear differential equations.
In order to do that, as was explained in the previous section, we need to assume
the field of constants is the complex field. Therefore, we have to extend ourselves
to the complex analytical category, i.e., all equations are complex analytical and
defined over complex analytical spaces.
For complex analytical Hamiltonian systems the General Principle works well
and we obtained the following result, which in some sense may be considered as
a generalization of a result by Ziglin in 1982 [148]. The essential idea is to con-
sider in the General Principle not only integrability of the variational equations
(characterized by the solvability of the identity component of its Galois group)
but commutativity of the identity component of the Galois group of the variational
equations. This is natural because, for integrable Hamiltonian systems, we have an
abelian Poisson Lie algebra of first integrals of maximal dimension.
Let H be a complex analytical Hamiltonian function defined on a symplectic
manifold M of (complex) dimension 2n and let XH be the Hamiltonian system de-
fined by H. In canonical coordinates, z = (x1 , ..., xn , y1 , ..., yn ), it is given classically
by
149
9
∂H
ẋi = ,
∂yi
∂H
ẏi = − ,
∂xi
i = 1, ..., n.
We recall here the definition of integrability for Hamiltonian systems. One
says that XH = (∂H/∂yi , ∂H/∂xi ) i = 1, ..., n, is completely integrable or Liouville
integrable if there are n functions f1 = H, f2 ,..., fn , such that
(1) they are functionally independent i.e., the 1-forms dfi i = 1, 2, ..., n, are
linearly independent over a dense open set U ⊂ M , Ū = M ;
(2) they form an involutive set, {fi , fj } = 0, i, j = 1, 2, ..., n.
We recall that in canonical coordinates the Poisson bracket has the classical
expression
n
∂f ∂g ∂f ∂g
{f, g} = − .
i=1
∂yi ∂xi ∂xi ∂yi
We remark that in virtue of item (2) above the functions fi , i = 1, ..., n are first
integrals of XH . It is very important to be precise regarding the degree of regularity
of these first integrals. In our contribution we assume that the first integrals are
meromorphic. Unless otherwise stated, this is the only type of integrability of
Hamiltonian systems that we consider in the next pages. Sometimes, to recall this
fact we shall talk about meromorphic (complete) integrability.
Now we can write the variational equations along a particular integral curve
z = φ(t) of the vector field XH

(2.4) ξ̇ = XH (φ(t))ξ.
Using the linear first integral dH(z(t)) of the variational equation it is possible
to reduce this variational equation and to obtain the so-called normal variational
equation which, in suitable coordinates, can be written as a linear Hamiltonian
system
η̇ = JS(t)η,
where, as usual,

0 I
J=
−I 0
is the standard matrix of the symplectic form of dimension 2(n − 1).
More generally, if, including the Hamiltonian, there are m meromorphic first
integrals independent over Γ and in involution, we can reduce the number of degrees
of freedom of the variational equation (2.4) by m and obtain the normal variational
equation (NVE) which, in suitable coordinates, can be written as a 2(n − m)-
dimensional linear system
(2.5) η̇ = JS(t)η,
152
150
where now J is the matrix of the symplectic form of dimension 2(n − m). For more
details about the reduction to the NVE, see [97] (or [95]).
Theorem 2.7 ([97], see also [95]). Assume a complex analytic Hamiltonian
system is meromorphically completely integrable in a neighborhood of the integral
curve z = φ(t) . Then the identity components of the Galois groups of the varia-
tional equations (2.4) and of the normal variational equations (2.5) are commutative
groups.
We remark that it is a typical version of several possible theorems. In some cases
it is interesting to add to the manifold M some points at infinity; thus we suppose
that we are in the following situation: M is an open subset of a complex manifold
M , M \ M is an hypersurface (which is by definition the hypersurface at infinity),
the two-form ω on M defining the symplectic structure extends meromorphically
on M and the vector field XH extends meromorphically on M . In such a case,
when (2.4) has irregular singular points at infinity, we only obtain obstructions to
the existence of first integrals which are meromorphic along Γ, ie, also at the points
at infinity of Γ; for example, for rational first integrals when M is a projective
manifold. From a dynamical point of view, the singular points of the variational
equation (2.4), Γ − Γ, correspond to equilibrium points, meromorphic singularities
of the Hamiltonian field or points at infinity.
Theorem 2.8. ([97], see also [95]) Consider a complex symplectic manifold
(M, ω), which is an open subset of a complex manifold M , M \ M being an hy-
persurface and ω admitting a meromorphic extension on M . Let XH be a mero-
morphic vector field on M which is analytic and Hamiltonian on M . If the system
ż = XH (z) is meromorphically integrable in a neighborhood in M of some integral
curve Γ with first integrals which extend into meromorphic functions on a neigh-
borhood of Γ, then the identity component of the Galois groups of (2.4) and (2.5)
(interpreted as differential equations on Γ) are commutative groups. In particular,
let M be an open domain of a symplectic complex space and assume the points at
infinity of (2.4) (or (2.5)) are irregular singular points and the identity component
of the Galois group of (2.4) (or (2.5)) is not commutative, then the Hamiltonian
system is not integrable by rational first integrals.
One of the essential points in the proof of the above theorems is the following
lemma:
Key Lemma: ([97], see also [95]) Let f be a meromorphic first integral of the
dynamical system (2.2). Then the Galois group of (2.3) has a non-trivial rational
invariant.
We remark that this Lemma is valid for general dynamical systems, not only
for Hamiltonian ones. Moreover, it is possible to generalize this lemma to tensor
invariants; for instance, to symplectic forms in the case of Hamiltonian systems or
to invariant volume forms. We shall not discuss these ideas here.
2.4. Extension to higher order variational equations. Theorem 2.7 (and

2.8) has been generalized to higher order variational equations VE k along Γ, with
k > 1 (the solutions of these equations are the quadratic, cubic, etc. contributions
to the flow of the Hamiltonian system along the particular solution z = φ(t) =
φ(z0 , t)), V E1 being the equation (2.3)[102].
151
11
The “fundamental” solution of VE k of a dynamical system (2.2) is given by

(φ(1) (t), φ(2) (t), . . . , φ(k) (t)),
where
1 (k)
φ(z, t) = φ(z0 , t) + φ(1) (t)(z − z0 ) + . . . + φ (t)(z − z0 )k + . . .
k!
the Taylor series up to order k of the flow φ(z, t) with respect to the variable z at the
∂k
point (z0 , t). That is, φ(k) (t) = k φ(z0 , t). The initial conditions are φ(1) (0) = Idm
∂z
and φ(j) (0) = 0 for all j > 1. We stress that, in contrast to some definitions, we
do not consider the differential equations for φ(k) , but for (φ(1) , φ(2) , . . . , φ(k) ), as
variational equation of order k.
The variational equation VE k is not linear, but it is in fact equivalent to a linear
differential equation: there exists a linear differential equation LVE k with coeffi-
cients in the field of meromorphic functions over Γ such that the differential exten-
sions generated by the solutions of VE k coincide with the Picard-Vessiot extensions
of LVE k . Then we can consider the Galois group Gk of VE k , Gk = Gal(VE k ),
i.e., of the LVE k . For simplicity we shall denote both VE k and LVE k by VE k .
Furthermore, the singular points of the equations VE k are the same as the first
order variational equation VE 1 and a singular point of VE k (k ≥ 1) is irregular if
and only if it is irregular for VE 1 .
Although in order to obtain the main theoretical results it is convenient to work
with the jet formalism (see [102]), from a computational practical point of view,
the higher order variational equations can be obtained using the small parameter
method of the masters (Poincaré, Liapunov,...). Hence, we expand the general
solution of the non-linear equation (2.2) along the particular solution φ(t)
ξ (2) 2 ξ (k) k
z(t) − φ(t) = ξ (1) ε + ε + ··· + ε + ··· ,
2! k!
being ε a small parameter. Introducing the above in equation (2.2) and equating
the same powers of ε, using the fact that
1 (2)
X(z) = X(φ(t)) + X (φ(t))(z(t) − φ(t)) + X (φ(t))(z(t) − φ(t))2 + · · · +
2!
1 (k)
X (φ(t))(z(t) − φ(t))k + · · · ,
k!
we obtain the variational equation of order k, VE k ,
(2.6) ξ̇ (j) = X (φ(t))ξ (j) + Pj (ξ (1) , . . . , ξ (j−1) ), j = 1, . . . , k,

where P1 ≡ 0 (i.e., VE 1 becomes (2.3)) and Pj is polynomial of degree j in
ξ (1) , . . . , ξ (j−1) , with meromorphic coefficients over Γ (i.e., in the same differential
field of coefficients of the first order variational equation). The initial conditions of
(1) (2) (k)
(2.6) are (ξ0 , ξ0 , ..., ξ0 ) = (Idm , 0, ..., 0). We observe that the solution of (2.6)
is obtained from the solution of the first order variational equation by successive
applications of the constants variation method of Lagrange: we substitute the so-
lutions of VE 1 in P2 (ξ1 ), we solve the second of the equations of VE 2 , i.e., the
(2)
equation in ξ (2) (with initial conditions ξ0 = 0) and we substitute all the above
154
152
solutions of VE 2 in P3 (ξ1 , ξ2 ), and so on. In particular, (2.6) is integrable if, and

only if, VE 1 is integrable.
For instance, the third order variational equation, VE 3 , is explicitly given by
the system
(2.7)
d
dt ξj,k = i ∂i Xj ξi,k ,
d
dt ξj,k1 k2 = ∂i Xj ξi,k1 k2 + i1 ,i2 ∂i21 ,i2 Xj ξi1 ,k1 ξi2 ,k2 ,
d
i
dt ξj,k1 k2 k3 = ∂X ξ + i1 ,i2 ∂i21 ,i2 Xj ξi1 ,k1 k2 ξi2 ,k3 +
i i j2 i,k1 k2 k3
∂ X ξ ξ + i1 ,i2 ∂i21 ,i2 Xj ξi1 ,k1 ξi2 ,k2 k3 +
i1 ,i2 i1 ,i32 j i1 ,k1 k3 i2 ,k2
i1 ,i2 ,i3 ∂i1 ,i2 ,i3 Xj ξi1 ,k1 ξi2 ,k2 ξi3 ,k3 ,
∂
where ξ (1) = (ξj,k ), ξ (2) = (ξj,k1 k2 ), ξ (3) = (ξj,k1 k2 k3 ), ξj,k := ∂zk φi (z0 , t), ξj,k1 k2 :=
2 3
∂ ∂
∂zk1 ∂zk2 φi (z0 , t), ξj,k1 k2 k3 := φi (z0 , t), j = k = k1 = k2 = k3 = 1, ..., m.
∂zk1 ∂zk2 ∂zk3
We remark that the matrix (ξj,k ) is in fact the fundamental matrix of the equation
(2.4).
Now we are going to describe the practical method of linearization of the equa-
tions VE k . The problem is to find a system of linear equations for
(ξ (1) (t), ξ (2) (t), ..., ξ (k) (t))
equivalent to VE k . It is enough to write the equations satisfied by the monomials
appearing in Pj . This is the content of the next lemma.
Lemma 2.9. Let z ∈ Cq . Assume the components (z1 , . . . , zq ) of z satisfy linear
q
homogeneous differential equations żi = j=1 aij (t)zj . Then the monomials z k :=
q ki
i=1 zi of order |k| = k1 + · · · + kq satisfy also a system of linear homogeneous
differential equations.
Proof. Let k = (k1 , . . . , kq ) a multi-index of non-negative integers. Then
⎛ ⎞
q q q
d k ⎝ kj −1
(2.8) z = kj zj ajr zr ziki ⎠ ,
dt j=1 r=1 i=1,i=j
the right hand side being also homogeneous of degree |k| in z.

We observe that the above Lemma is nothing other than the pull–back to
k q
the symmetric fibre bundle, qS (C ), of the connection associated to the linear
differential equation żi = j=1 aij (t)zj , connected to the Tannakian formalism
(see [95], Section 2.3).
Then, in order to linearize VE k , after the last equation corresponding to VE k
we can supplement the system of linear differential equations with the equations
for the components of (ξ (i1 ) )m1 , (ξ (i2 ) )m2 , . . . , (ξ (is ) )ms . For more details about
this linearization see our original paper, [102].
Since we can now consider the equations VE k as linear differential equations,
we can talk about their Picard-Vessiot extensions and about their Galois groups
Gk . Then it was conjectured in [95] (Section 8.3) that a necessary condition for
meromorphic integrability is that the identity component of Gk must be commuta-
tive, for any k. This can be interpreted as some generalized version of our General
Principle to this context. So, using also a generalized version of the Key Lemma
153
13
to Hamiltonian dynamical systems, in a joint work of the authors with Simó we

proved the following theorem.
Theorem 2.10 ([102]). Assume that a complex analytical Hamiltonian system
is integrable by meromorphic first integrals in a neighborhood of the integral curve
z = φ(t). Then the identity components (Gk )0 , k ≥ 1, of the Galois groups of the
variational equations along Γ are commutative.
Example: a Hénon-Heiles system. Let us to consider the two degrees of free-

dom Hamiltonian system defined by the cubic Hamiltonian
1 1 1 1 1
(2.9) H = (y12 + y22 ) + x21 + x22 + x31 + x1 x22 .
2 2 2 3 2
In [102] it was proven non-integrable, we follow this reference. In order to prove
this we use the method in Appendix A, A.3, a).
This system is one of the Hénon-Heiles family of one-parameter Hamiltonians
considered by Ito [50]. By means of the first order variational equation it was
proven that for all but four of the values of the parameter, the systems in this family
are non-integrable, see [50, 94, 95]. Three of these remaining cases are trivially
integrable. The fourth case is (2.9) and its non-integrability was conjectured from
numerical experiments. We can prove the non-integrability of this last case using
Theorem 2.10. As we will see, we obtain an obstruction given by a non trivial
residue at the integrand of a quadrature appearing in the solution of the third
variational equation.
For the Hamiltonian (2.9), the plane x2 = y2 = 0 is invariant, foliated by
integral curves Γh which are generically elliptic curves
1 2 1 2 1 3
y + x + x = h,
2 1 2 1 3 1
parametrized in time by x1 = −6℘(t) − 12 , y1 = ẋ1 = −6℘(t), ˙ x2 = y2 = 0, where
h
℘ is the Weiertrass elliptic function with invariants g2 = 12 1 1
and g3 = 108 − 18 . In
the above computations we used the differential equation satisfied by the function
℘(t): ℘˙ 2 = 4℘3 − g2 ℘ − g3 , see Appendix C.
For h∗ = 1/6, the elliptic curve degenerates to a rational one, Γh∗ := Γ: the
real period of x1 (t) and y1 (t) goes to infinity, and Γ is parametrized in time by
3/2 −(3/2) sinh(t/2)
(2.10) x1 (t) = − 1, y1 (t) = , x2 = y2 = 0.
cosh2 (t/2) cosh3 (t/2)
We observe that in any case for Γh , h arbitrary, x1 (t) and y1 (t), have only one pole
in a fundamental domain in the complex plane. This pole will be a singular point
of the corresponding variational equations.
Then the first variational equation VE 1 along Γh = Γ is given by
ξ¨1 = 12 ℘(t)ξ1 ,
(2.11)
ξ¨2 = 6 ℘(t) − 12 ξ2 .
(in fact, we are interested in the fundamental matrix (ξi,j ) i, j = 1, .., 4 of the above
system). The first of the equations in (2.11) is the tangential variational equation
and the second one is the normal variational equation, with coefficients −1 − 2x1 (t)
and −1 − x1 (t), respectively. We know that the tangential variational equation has
156
154
a solution belonging to the field of elliptic functions K = C(℘(t), ℘(t)), ˙ since a

particular solution of the tangential variational equation is obtained by derivation
of the particular solution z = z(t) along which we compute the variational equation
and thus, in this case, ξ1 = ẋ1 = y1 = 6℘(t).
˙ In fact a direct computation verifies
that ℘(t)
˙ is a particular solution, where again we use the differential equation
satisfied by ℘(t). Thus, the first of the equations in (2.11) falls in the Lamé case,
see Appendix C. In an analogous way, the normal variational equation has also a
particular solution ξ2 = ℘(t) + 12 1
∈ K. Hence, both equations in (2.11) fall in the
Lamé case and the Galois group of both of them is not finite (Appendix C).
From the results in Appendix A, A.3, a), Proposition A.6, we know that if a
particular solution of VE k , with k > 1, has a solution with a local logarithm, then
the system is not integrable. We will prove this is the case for VE 3 , i.e., that some
of the final elements ξj,k1 k2 k3 are not meromorphic. We select the ξ2,222 . Instead of
working with the family of curves Γh , h = h∗ , with VE 1 given by (2.11), we shall
work with the integral curve Γ. Then it is clear that, from the analytic dependence
on the parameter h, if we have a non-trivial monodromy (i.e., a local logarithm)
around the singular point in the variational equation VE 3 over the integral curve Γ,
then the variational equation over Γh also has a non-trivial monodromy for h = h∗
and small |h − h∗ |.
Let a31 = −1 − 2x1 , a42 = −1 − x1 the only non-zero and non-trivial elements
in the matrix of coefficients of V E1 along Γ. To obtain ξ2,222 from the equation
(2.7), we only need to integrate the following systems

ξ˙2,2 0 1 ξ2,2 ξ˙1,22 0 1 ξ1,22 0
(2.12) = , = +
ξ˙4,2 a42 0 ξ4,2 ξ˙3,22 a31 0 ξ3,22 −ξ2,2
2
and

ξ̇2,222 0 1 ξ2,222 0
(2.13) = + .
ξ̇4,222 a42 0 ξ4,222 −3ξ2,2 ξ1,22
It is clear that to integrate (2.12) and (2.13) we need to solve the first order varia-
tional equations, both tangential and normal which, as above for Γh (equation(2.11)),
are uncoupled. The solutions can be written explicitly. To shorten the notation we
introduce c := cosh(t/2) and s := sinh(t/2). Then
15ts 15 5 c2 4
ξ1,1 = − 3
+ 2− − , ξ1,3 = − y1 ,
16c 8c 8 4 3
15t(3 − 2c2 ) 45s sc 4
ξ3,1 = − − − , ξ3,3 = (x1 + x21 ),
(2.14) 32c4 16c3 4 3
tx1 3s
ξ2,2 = 2x1 , ξ2,4 = + ,
2 2c
x1 ty1 3
ξ4,2 = 2y1 , ξ4,4 = + + 2.
2 2 4c
Furthermore

2 16 3
(2.15) ξ1,22 = ξ1,1 − x1 + ξ1,3 K(t),
8 9
where

45 45 15 45 15 3
K(t) = t − + 4 − 2 +s − 5 + 3 − +c .
16c6 8c 4c 8c 2c c
155
15
We remark that one of the columns of the fundamental matrix of the normal vari-
ational equations coincides (except by a factor of 2) with (2.10). This is true for
any h because (x1 , y1 ) are solutions of the first equation in (2.12).
Having (2.14) and (2.15) we are ready to solve (2.13). As the homogeneous
part coincides with the first order normal variational equation, the solution, after
closing the loop, is given by the method of variation of constants

ξ2,222 ξ2,2 ξ2,4 −ξ2,4 R dt
(2.16) = ,
ξ4,222 ξ4,2 x4,4 γ ξ2,2 R dt
where R(t) = −3ξ2,2 ξ1,22 . It is readily checked that the residues inside the integral
72
are 72/5 and 0, respectively. Hence, the final value of ξ2,222 after the loop is 2π i,
5
due to the existence of a local logarithmic term. Hence, from Proposition A.6 of
Appendix A, we have proven the following result.
Proposition 2.11 ([102]). The system (2.9) is not integrable by means of
meromorphic first integrals.
We observe that in order to prove there is non–trivial local monodromy in ξ2,222 ,
it is also possible to use another method: instead of using the analytic dependence
with respect to h, we can work directly with the variational equations along Γh with
some h = 0 and in an analogous way to check the existence a non-trivial residue in
the integrand ξ2,4 R. This is the standard method followed by several authors.
We remark that since the Galois group G3 is connected (i.e., G3 = (G3 )0 ), it is
possible to use the weaker result given by Lemma A.4 of Appendix A rather than
Proposition A.6.
Recently a general methodology to deal with non-integrability criteria using
higher variational equations was developed by R. Martı́nez and C. Simó [87, 88,
89]. Using this methodology, they prove in particular the non-integrability of a non-
linear spring pendulum problem and the non-integrability of the Swinging Atwood
Machine for the values of the parameter that, in each case, cannot be decided
using first order variational equations ([78]). For the Swinging Atwood Machine
with pulleys it is possible to prove non-integrability in all cases using only first
variational equations [115].
As a conclusion to this section, we can say that all of our approach is based
upon two simple facts:
(i) A heuristic guiding General Principle.
(ii) The Key Lemma.
3. Homogeneous Potentials and Related Problems

In this section we survey how the preceding methods have been applied to
classical mechanics systems involving homogenous potentials. Some other related
problems, like Celestial Mechanical ones, are included.
3.1. Non-integrability of Homogeneous Potentials . Here we recall the

main general non-integrability result about Hamiltonians with homogeneous poten-
tials obtained by the authors some years ago [98].
Consider an n-degrees-of-freedom Hamiltonian system with Hamiltonian
158
156
1 2
(3.1) H(x, y) = T + V = (y + ... + yn2 ) + V (x1 , ..., xn ),
2 1
V being a complex homogeneous function of integer degree k and 2 ≤ n.
From the homogeneity of V , it is possible to obtain an invariant plane
x = z(t)c,
y = ż(t)c,
where z = z(t) is a solution of the (scalar) hyperelliptic differential equation
2
(1 − z k )
ż 2 =
k
0), and c = (c1 , c2 , ...cn ) is a solution of the equation
(where we assume case k =
(3.2) c = V (c).
This is our particular solution Γ along which we compute the variational equation
VE and the normal variational equation NVE. We shall call these the homothetical
solutions of the Hamiltonian system (3.1) and denote solutions of (3.2) as homoth-
etical points. In most of the references about the integrability of the homogeneous
potentials the solutions of (3.2) are called Darboux points (see [75], for instance);
we use the standard terminology in the Celestial Mechanical case (see later).
The VE along Γ is given in the temporal parametrization by
η̈ = −z(t)k−2 V (c)η.
Assume V (c) is diagonalizable. Due to the symmetry of the Hessian matrix V (c),
it is possible to express the VE as a direct sum of second order equations
η̈i = −z(t)k−2 λi ηi , i = 1, 2, ..., n,

where we keep η for the new variable, λi being the eigenvalues of the matrix V (c).
We call these eigenvalues Yoshida coefficients. One of the above second order
equations is the tangential variational equation, say, the equation corresponding to
λn = k − 1. This equation is trivially solvable, whereas the NVE is an equation in
the variables ξ := (η1 , ..., ηn−1 ) := (ξ1 , ..., ξn−1 ), i.e.,
ξ¨ = −z(t)k−2 diag(λ1 , ..., λn−1 )ξ.

Now, following Yoshida [146], we consider the change of variable (which hap-
pens to be a finite branched covering map),
Γ → P1 ,
k
given by t → x, where x =: z(t) (here Γ is the compact hyperelliptic Riemann
surface of the hyperelliptic curve w2 = k2 (1 − z k ), see [98] or [95] for the notation
and technical details). Thanks to the symmetries of this problem, we obtain as
NVE a system of independent hypergeometric differential equations in the new
independent variable x
d2 ξ k − 1 3k − 2 dξ λi
(ANVEi ) x(1 − x) +( − x) + ξ = 0, i = 1, 2, ..., n − 1.
dx2 k 2k dx 2k
157
17
Each of these equations (ANVEi ), corresponding to the Yoshida coefficient

λi , is part of the system called the algebraic normal variational equation ANVE
(see Appendix A, Section A.2). In fact, the ANVE splits into a system of n − 1
independent equations (ANVEi ), i = 1, ..., n − 1. Then it is clear that the ANVE
is integrable if, and only if, each of the (ANVEi ) is also integrable. That this,
the identity component of the Galois Group of the ANVE is solvable if, and only
if, each one of the identity components of the Galois Group of the (ANVEi ) i =
1, 2, . . . , n − 1, is solvable.
As was observed by Yoshida, each one of the above (ANVEi ) is an hypergeo-
metric equation (equation (B.2) of Appendix B) with three regular singular points
at x = 0, x = 1 and x = ∞. By Theorem A.1 of Appendix A the identity compo-
nent of the Galois Group of the NVE is the same as the identity component of the
Galois Group of the ANVE. By adapting Kimura’s table (Theorem B.1 of Appendix
B) of integrable hypergeometric equations to the new hypothesis, namely assuming
that the Galois differential group of each of the variational equations must have a
commutative identity component, we obtain the following result (in fact, for this
particular family of hypergeometric equations all the cases of integrability have a
commutative identity component):
Theorem 3.1 ([98], see also [95]). Let XH be a Hamiltonian system given
by (3.1) and c an homothetical point such that V (c) is diagonalizable. If XH is
meromorphically completely integrable, then each pair (k, λi ) matches one of the
following items (p being an arbitrary integer):
k λ k λ
2
1 k p + p (p − 1) k2 10 −3 25
24 − 1
24
12
5 + 6p
2 2 arbitrary z ∈ C 11 3 − 24
1
+ 1
24(2 + 6p)2
2
3 −2 arbitrary z ∈ C 12 3 − 24
1
+ 241 3
2 + 6p
2 2
4 −5 49
40 − 1
40
10
3 + 10p 13 3 − 24
1
+ 241 6
5 + 6p
(3.3)
2 2
5 −5 49
40 − 1
40 (4 + 10p) 14 3 − 24
1 1
+ 24 12
5 + 6p
2 2
6 −4 9
8 − 1
8
4
3 + 4p 15 4 − 18 + 18 43 + 4p
2
7 −3 25
− 24
1
(2 + 6p)2 16 5 − 40
9 1
+ 40 10
+ 10p
24 2 3
2
8 −3 24 − 24
25 1 3
2 + 6p 17 5 − 40
9 1
+ 40 (4 + 10p)
2 1 k−1

9 −3 25
24 − 1
24
6
5 + 6p 18 k 2 k + p (p + 1) k
This theorem is a generalization of a necessary condition of integrability ob-

tained by Yoshida using Ziglin’s approach [146].
Hence, in order to prove the non-integrability of a given Hamiltonian system
with a homogeneous potential:
(i) we find the homothetical points, solutions ci of the equation
c = V (c)
(ii) we prove that for some of the ci in (i), at least one of the eigenvalues of
of V (ci ) is not inside the table (3.1).
160
158
Furthermore, we can stretch the above result a bit further thanks to the results
in [103]. If XH has p first integrals f1 = H, . . . , fp in involution and independent
on Γ, including the Hamiltonian we have a set of m eigenvalues (k − 1 among them,
corresponding to H) that belong to Table (3.3) and the normal variational equa-
tions, NVE, are now n − p of the initial variational equations. Reordering indexes if
needed, let us write them as VEp+1 , . . . , VEn with corresponding differential Galois
groups Gp+1 , . . . , Gn and let us write the eigenvalues corresponding to f1 , . . . , fp
as λ1 = k − 1, . . . , λp in Table (3.3). Then, if there is an additional first integral
independent of the set {f1 , . . . , fp }, the Galois group of the normal variational equa-
tions must necessarily possess a rational invariant. Recently A. J. Maciejewski, M.
Przybylska and H. Yoshida proved the following:
Theorem 3.2 ([79]). Let XH be a Hamiltonian field given by (3.1). If there
is at least an additional single first integral f independent with {f1 , . . . , fp } on a
neighborhood of Γ (but may be dependent on Γ), then we have one of the following
two situations:
1. At least one of the eigenvalues λ1 , . . . , λn−p belongs to Table (3.3).
2. There are 1 ≤ i < j ≤ n − p such that

(k − 2)2 + 8kλi − (k − 2)2 + 8kλj ∈ 2kZ.
We will actually perform a step further and, as a by-product, obtain an alter-

native proof for Theorem 3.2.
Theorem 3.3 ([103]). Let XH be a Hamiltonian field given by (3.1). If there
is at least an additional single first integral f independent with {f1 , . . . , fp } on a
neighborhood of Γ (but may be dependent on Γ), then we have one of the following
two situations:
1. At least one of the eigenvalues λ1 , . . . , λn−p belongs to Table (3.3).
2. There exist 1 ≤ i < j ≤ n − p such that

(3.4) (k − 2)2 + 8kλi − (k − 2)2 + 8kλj ∈ 2kZ.
Moreover if we divide the set of eigenvalues {λ1 , . . . , λn−p } in equivalence
classes, Λ1 = {λ1,1 , . . . , λ1,k1 }, . . . Λr = {λr,1 , . . . , λr,kr }, {λK+1 }, . . . , {λn−p },
with respect to the relation defined by (3.4) with k1 , k2 , . . . , kr all greater
than
r 1 (by reordering the eigenvalues we can assume this) and K :=
i=1 ki . Then XH can have at most 2K − 3r additional meromorphic
first integrals.
There is another extension of Theorem 3.1 to non-homogeneous potentials.
Given a potential which is not homogenous, but is expressed as a finite sum of
homogeneous potentials
(3.5) V = V k 1 + . . . Vk m ,
being km the degree of homogeneity of the corresponding term, ki < ki+1 , it is
possible to apply Theorem 3.1 by using a transformation that traces back to the
Levi-Civita regularization. Therefore, the transformed system depends on a pa-
rameter , which is essentially the inverse of the energy level. For the limit cases
= 0, ∞ we obtain systems with homogeneous potential Vk1 , Vkm . Then Mondéjar,
159
19
using a previous result about parametric Hamiltonian systems ([91]), was able to
prove the following result.
Theorem 3.4 ([92]). If the Hamiltonian system with potential (3.5) is com-
pletely integrable with meromorphic first integrals, then both potentials Vk1 and Vkm
must satisfy the conditions (1)-(18) of Theorem 3.1.
It is worth noting that, in some sense, the above theorem generalizes previous
results for two–degrees–of–freedom systems obtained by Hietarinta and Yoshida
[44, 147].
3.2. Homogenous Polynomial Potentials. A natural problem is to apply

the table (3.3) to the study of the integrability of Hamiltonian systems associated
to polynomial potentials. So, in this section we assume that the potential in (3.1) is
polynomial of degree k. Although this problem has already been studied by several
authors such as Maciejewski, Nakagawa, Yoshida and Przybylska, it remains a very
active area of research. The final ambitious goal is to obtain a complete classification
of the integrable potentials. The strategy is as follows:
(i) applying the results of the above section; this gives us some candidates to
integrability,
(ii) trying a direct integrability proof for the above candidates for integrability.
The classification is completed for n = 2 (two–degrees of freedom) and k = 3; it
is almost completed for n = 2 and k = 4 (we remark that the cases k = 1, 2 are
trivially integrable). Now we restrict to two-degrees of freedom.
In 2001–2002 Nakagawa and Yoshida obtained the list of all two– degrees of
freedom integrable homogeneous polynomial potentials such that the additional first
integral is polynomial of degree four in the momenta [105, 106]. In the proof they
used Ziglin’s lemma ([148]) around the homothetical solutions (i.e., something very
related to the Key Lemma in Section 2.1) and a previous study by Hietarinta by
means of the direct method, (ii) above [43, 44]. Then Maciejewski and Przybylska
studied the meromorphic non-integrability for k = 3, 4, [72, 75]. We follow [75].
If we restrict table (3.3) to natural numbers we obtain the following necessary
conditions for integrability.
k λ
1 k p + p (p − 1) k2
2
2 3 − 24
1
+ 241
(2 + 6p)
2
3 3 − 24
1
+ 241 3
2
+ 6p
2
4 3 − 24
1
+ 241 6
5
+ 6p
(3.6) 2
5 3 − 24
1 1
+ 24 12
5 + 6p
2
6 4 − 18 + 18 43 + 4p
2
7 5 − 40
9 1
+ 40 10
3 + 10p
2
8 5 − 40
9 1
+ 40 (4 + 10p)
1 k−1

9 k 2 k + p (p + 1) k
162
160
where we do not consider the quadratic potentials, since all of them are integrable,
i.e., from now on in this section k > 2.
For n = 2, we have only two eigenvalues λ1 = k − 1, λ2 of V (ci ), at a
homothetical point c = (c1 , c2 ). Denoting the non-trivial eigenvalue λ2 := λ,
by means of a detailed algebraic analysis and using the residue theorem over the
Riemann sphere along a suitable differential form, the following remarkable global
universal relation between the several λ’s for a given degree of the potential is
proven.
Theorem 3.5 ([75]). Assume the polynomial homogeneous potential V (x1 , x2 )
has k different homothetical points c1 , ..., ck with corresponding non-trivial eigen-
values λ1 ,..., λk . Then,
k
1
(3.7) = −1.
i=1
λi − 1
For example, if we apply the above theorem, as well as table (3.6), to potentials
of degree k = 4 with four homothetical points, it is shown in [75] that the only
possible cases for {λ1 , λ2 , λ3 , λ4 } are as in table (3.8).
1 {0, 0, 3, 3}
2 { 38 , 6, 6, 6}
(3.8) 3 { 38 , 3, 21, 21}
4 { 38 , 35
8
, 35
8
, 136}
5 { 38 , 3, 15, 36}
It is worth pointing out that, for generic homogeneous polynomial potentials,
the assumptions of Theorem 3.5 are satisfied, since for two–degrees of freedom, the
potential is defined in a natural way over the Riemann sphere P1 , and the exis-
tence of homothetical points is reduced to the search of solutions of some suitable
polynomials in one single variable ([75]).
Then, using our table (3.6) and Theorem 3.5, the authors obtained that under
the above assumptions the number of integrable potentials of a given degree must
be finite:
Theorem 3.6 ([75]). For a given degree k, the family of inequivalent homoge-
nous integrable polynomials which satisfy the assumption of Theorem 3.5 is finite.
For non-generic potentials of degree k the number of homothetical points is less
than k, but for k = 3 and k = 4 it is possible to (nearly) finish the classification and
reconstruct the possible integrable potentials; in particular, for some non-generic
families it is possible to generalize Theorem 3.5 in a suitable way. So, using some
constructive methods of algebraic geometry and an analysis of case by case it is
shown in [72] that for k = 3 there are no other integrable cases that those seen
already in references [43, 44]. For k = 4 the problem is more difficult, and the
classification of the integrable cases is not complete: integrability remains open for
the discrete infinite family of potentials
1−α 2 1
(3.9) V (x1 , x2 ) = x1 (x1 + ix2 )2 + (x21 + x22 )2 ,
2 4
161
21
where the parameter α takes values in some discrete set (obtained by means of
the table (3.6)). It is worth remarking that in order to study the integrability of
the generic family given by table (3.8) (and a similar, simpler table for k = 3), an
essential tool was the use of the higher variational equations, i.e., Proposition A.6
of Appendix A.
For more than two degrees of freedom it is possible to generalize some of the
above results. In [116], the author was able to prove, for arbitrary n > 2, the
finiteness of the number of integrable homogeneous potentials of a fixed degree k
with a maximal number of homothetical points. A new method enters in the proof:
the Kovalewskaya exponents of an auxiliary gradient system in the configuration
space with the field given by the gradient of the potential. The equilibrium points
of the above gradient system are the homothetical points of the potential. It is
easy to prove that the Kowaleskaya exponents at the homothetical point Λ1 , ..., Λn
coincide with shifted eigenvalues of the Hessian of the potential at this point c,
λi = Λi + 1. Then a universal global relation which generalizes the equation (3.7)
is studied for potentials with a maximal number of homothetical points. From this
relation the author obtains the finiteness of the integrable potentials. See [116] for
details.
3.3. Some Rational Potentials . In a joint work of Simó with the first
author the integrability of families of two-degrees of freedom potentials with an
invariant plane and normal variational equations of Lamé type ([94] it was studied,
see also [95]). Under suitable assumptions of regularity, it is easy to see that if the
invariant plane is given by x2 = y2 = 0, the potential should be of the form
x22
(3.10) V (x1 , x2 ) = φ(x1 ) − α(x1 )
+ β(x1 , x2 )x32 .
2
The NVE associated to any integral curve lying on the invariant plane is
(3.11) ξ¨ = α(x1 (t))ξ.
It is clear that the first problem is to find the families of potentials with a
given (3.11). In the recent paper [3], assuming that (3.11) can be expressed with
polynomial coefficients, the authors completely solved this problem as stated in
[94] in an algorithmic way and applied it to the integrability of several families of
rational potentials. Here we review these results in [3].
From now on, we will write a(t) = α(x1 (t)), for a generic curve z = z(t) =
(x1 (t), y1 (t)) lying on the invariant plane and parameterized by t. Then, the NVE
is written
(3.12) ξ¨ = a(t)ξ.
3.3.1. The determination of the potentials. Problem. Assume that a(t) is a
root of a given differential polynomial Q(a, ȧ, ä, . . .) ∈ C[a, ȧ, ä, . . .]. We want to
compute all potentials in (3.10) satifiying such a condition.
So, we shall give a method to compute, for any given Q(a, ȧ, . . .), the family of
potentials with invariant plane x2 = y2 = 0 such that, for any integral curve lying
on this invariant plane, the coefficient a(t) of the NVE satisfies,
(3.13) Q(a, ȧ, ä, . . .) = 0,
by solving certain differential equations.
164
162
We should notice that, for a generic integral curve z(t) = (x1 (t), y1 (t)), y1 (t) =
ẋ1 (t), lying on x2 = y2 = 0, equation (3.12) depends only on the values of functions
α and φ. It depends on α(x1 ), since a(t) = α(x1 (t)). We observe that the curve
z(t) is a solution of the restricted Hamiltonian,
y12
(3.14) + φ(x1 ) h=
2
whose associated Hamiltonian vector field is,
∂ dφ ∂
(3.15) Xh = y1 − ,
∂x1 dx1 ∂y1
dφ
thus x1 (t) is a solution of the differential equation, ẍ1 = − dx1
, and then, the
relation of x1 (t) is given by φ.
Since z(t) is an integral curve of Xh , for any function f (x1 , y1 ) defined on the
invariant plane x2 = y2 = 0 we have
d ∗
z (f ) = z ∗ (Xh f ),
dt
where z ∗ denotes the usual pull–back of functions. Then, using a(t) = z ∗ (α), we
have for each k ≥ 0,
dk a
(3.16) = z ∗ (Xhk α),
dtk
so that,
Q(a, ȧ, ä, . . .) = Q(z ∗ (α), z ∗ (Xh α),∗ (Xh2 α), . . .).
There is an integral curve of the Hamiltonian through each point of x2 = y2 = 0,
and thus we have the following.
Proposition 3.7. Let H be a Hamiltonian of the family (3.10), and Q(a, ȧ, ä, . . .)
a differential polynomial with constant coefficients. Then, for each integral curve ly-
ing on x2 = y2 = 0, the coefficient a(t) of the NVE (3.12) satisfies Q(a, ȧ, ä, . . . , ) =
0, if and only if the function
Q̂(x1 , y1 ) = Q(α, Xh α, Xh2 α, . . .),
vanishes on x2 = y2 = 0.
Now we see that Q̂(x1 , y1 ) is a polynomial in y1 and its coefficients are differ-
ential polynomials in α, φ. So, if we write down the expressions for successive Lie
derivatives of α, we obtain
dα
(3.17) Xh α = y1 ,
dx1
d2 α dφ dα
(3.18) Xh2 α = y12 2 − ,
dx1 dx1 dxi

d3 α d dφ dα dφ d2 α
(3.19) Xh3 α = y13 − y1 +2 ,
dx31 dx1 dx1 dx1 dx1 dx21

d4 α d d dφ dα dφ d2 α d3 α dφ
Xh4 α = y14 4 − y12 +2 +3 3 +
dx1 dx1 dx1 dx1 dx1 dx1 dx21 dx1 dx1
163
23

d dφ dα dφ d2 α dφ
(3.20) + +2 .
dx1 dx1 dx1 dx1 dx21 dx1
In general form we have,
∂Xhn α dφ ∂Xhn
(3.21) Xhn+1 α = y1 − ,
dx1 dx1 ∂y1
and it inductively follows that they all are polynomial in y1 , its coefficients being
differential polynomials in α, φ. If we write it down explicitly,

(3.22) Xhn α = En,k (α, φ)y1k ,
n≥k≥0
r s

we can see that the coefficients En,k (α, φ) ∈ C α, φ, ddxαr , ddxφs , satisfy the following
1 1
recurrence law,
d dφ
(3.23) En+1,k (α, φ) = En,k−1 (α, φ) − (k + 1)En,k+1 (α, φ)
dx1 dx1
with initial conditions,
dα
(3.24) E1,1 (α, φ) = , E1,k (α, φ) = 0 ∀k = 1.
dx1
We observe that the recurrence law (3.23) and initial conditions (3.24) deter-
mine the coefficients En,k (α, φ). We can compute the value of some of them easily:
n
• En,n (α, φ) = ddxα
n for all n ≥ 1.
1
• En,k (α, φ) = 0 if n − k is odd, or k < 0, or k > n.
As an illustration of the above method we now compute families of potentials
(3.10) associated to a specific well–known NVE. Although, in order to perform
these computations, we need to solve polynomial differential equations, we will see
that we can deal with this in a series of cases. Particularly, when Q is a differential
linear operator, we will obtain equations involving products of few linear differential
operators.
Example 1: NVE of harmonic oscillator type. Harmonic oscillator equation
is
(3.25) ξ¨ = c0 ξ,
with c0 constant. Then, a Hamiltonian of type (3.10) gives such NVE if ȧ = 0.
dα
Looking at formula (3.17), it follows that dx1
= 0, so that α is a constant. We
conclude that the general form of a Hamiltonian (3.10) which gives rise to NVE of
the type (3.25) is,
y12 + y22
H= + φ(x1 ) + λ0 x22 + β(x1 , x2 )x32 ,
2
λ0 being a constant, and φ, β arbitrary analytical functions.
Example 2: NVE of Airy type. In [9], Audin notices that the Hamiltonian,
y12 + y22
+ x1 x22
2
gives an example of a simple non-integrable classical Hamiltonian, since its NVE
along any integral curve in the invariant plane x2 = y2 = 0 is an Airy equation. Here
166
164
we compute the family of classical Hamiltonians that have NVE of type Airy for
integral curves lying on the above invariant plane. General form of Airy equation
is
(3.26) ξ¨ = (c0 + c1 t)ξ
with c0 , c1 = 0 two constants. If follows that a Hamiltonian gives rise to NVE of
this type if ä = 0, and ȧ = 0. The equation ä = 0 gives, by Proposition 3.7 as we
see in formula (3.18), the following system:
d2 α dφ dα
(3.27) = 0, = 0.
dx21 dx1 dx1
It splits into two independent systems,
2
d α
dα dx21
=0
(3.28) = 0, dφ
dx1 dx1 = 0
Solutions of the first one fall into the previous case of harmonic oscillator. Then,
taking the general solution of the second system, we conclude that the general form
of a classical Hamiltonian of type (3.10) with Airy NVE is:
y12 + y22
(3.29) H= + λ0 + λ1 x22 + λ2 x1 x22 + β(x1 , x2 )x32 ,
2
with λ2 = 0.
Example 3: NVE of quantum harmonic oscillator type. Let us now consider
3 2
now equations with ddt3a = 0, and ddt2a = 0, it is
(3.30) ξ¨ = (c0 + c1 t + c2 t2 )ξ
with c2 = 0. Those equations can be reduced to a quantum harmonic oscillator
equation by an affine change of t. Using Proposition 3.7 and formula (3.19), we
obtain the following system of differential equations for α and φ:
d3 α dα d2 φ d2 α dφ
3 = 0, 2 +3 2 = 0.
dx1 dx1 dx1 dx1 dx1
The general solution of the first equation is
λ1 λ2 λ3
α= + x1 + x21 ,
2 2 2
and substituting it into the second equation we obtain a linear differential equation
for φ,
d2 φ 2λ3 dφ
2 +3 = 0,
dx1 λ2 + 2λ3 x1 dx1
this equation is integrated by two quadratures, and its general solution is
λ4
φ= + λ0 .
(λ2 + 2λ3 x1 )2
We conclude that the general formula for Hamiltonians of type (3.10) with NVE
(3.30) for any integral curve lying on x2 = y2 = 0 is
y12 + y22
H= +
2
λ4
(3.31) + λ0 − λ1 x22 − λ2 x1 x22 − λ3 x21 x22 + β(x1 , x2 )x32 ,
(λ2 + 2λ3 x1 )2
165
25
with λ3 = 0.
We observe that formula (3.31) yields non-linear dynamics in the invariant
plane x2 = y2 = 0. Notice that these dynamics are continuously deformed to linear
dynamics when λ4 tends to zero. In the general case, for a fixed energy h, we have
the general integral of the equation:
8λ23 h2 (t − t0 )2 = h(λ2 + 2λ3 x1 )2 − λ4 .
Example 4: NVE with polynomial coefficient a(t) of odd degree. Here

we generalize Example 2. Let us consider for n > 0 the following differential
polynomial,
dm a
Qm (a, ȧ, . . .) = m .
dt
It is obvious that a(t) is polynomial of degree n if and only if Qn (a, ȧ, . . .) = 0 and
Qn+1 (a, ȧ, . . .) = 0.
Looking a Proposition 3.7, we see that a Hamiltonian (3.10) has NVE along a
generic integral curve lying on x2 = y2 = 0,
(3.32) ξ¨ = Pn (t)ξ,
where Pn (t) polynomial of degree n, if and only if Xhn α = 0 and Xhn+1 α vanishes
on x2 = y2 = 0. Let us remind expression (3.22), Xhn+1 α vanish in x2 = y2 = 0 if
and only if (α, φ) is a solution of the differential system
Rn+1 = {En+1,0 (α, φ) = 0, . . . , En+1,n+1 (α, φ) = 0}.
A particular solution of Rn+1 which does not satisfy Rn , is given by φ = λ0 ,

α(x1 ) = Qn (x1 ), polynomial of degree n. Then the Hamiltonians,
y12 + y22
(3.33) H= + λ0 + Qn (x1 )x22 + β(x1 , x2 )x32 ,
2
have NVE, along a generic integral curve lying on x2 = y2 = 0, of the form (3.32).
If n is an even number, there are more solutions of the differential system
Rn+1 not verifying Rn , being a particular case the potentials with generic quantum
harmonic oscillators, computed above. We will prove, using the recurrence law
(3.23), that for odd n, the above family is the only solution of Rn+1 not verifying
Rn .
dφ
Lemma 3.8. Let (α, φ) be a solution of R2m . Then, if dx1
= 0, then (α, φ) is
a solution of R2m−1 .
Proof. As 2m − 1 is odd, E2m−1,2k (α, φ) = 0 for all m − 1 ≥ k ≥ 0. Then let
us prove that E2m−1,2k+1 (α, φ) = 0 for all m − 2 ≥ k ≥ 0.
In the first step of the recurrence law defining R2m ,
dE2m−1,1 dφ
0 = E2m,0 (α, φ) = (α, φ) − E2m−1,1 (α, φ),
dx1 dx1
dφ
we use dx1 = 0, and E2m−1,−1 (α, φ) = 0 to obtain
E2m−1,1 (α, φ) = 0.
If we assume E2m−1,2k+1 (φ, α) = 0, substituting it in the recurrence law
dE2m−1,2k dφ
E2m,2k+1 (α, φ) = (α, φ) − 2(k + 1) E2m−1,2(k+1) (α, φ),
dx1 dx1
168
166
we obtain that
E2m−1,2(k+1) (α, φ) = 0,
and we conclude by finite induction.
Corollary 3.9. Let H be a classical Hamiltonian of type (3.10). Then the
following statements are equivalent,
(1) The NVE for generic integral curve (3.12) lying on x2 = y2 = 0 has
polynomial coefficient a(t) of degree 2m − 1.
(2) H can be written
y12 + y22
(3.34) H= + λ0 − P2m−1 (x1 )x22 + β(x1 , x2 )x32 ,
2
for λ0 constant and P2m−1 (x1 ) polynomial of degree 2m − 1.
Proof. It is clear that condition 1 is satisfied if and only if (α, φ) is a solution
of R2m and it is not a solution of R2m−1 . By the previous Lemma, this implies
dφ d2m α
dx1 = 0, and the system R2m is thereby reduced to dx2m 1
= 0 and then, φ is
therefore a constant and α must be a polynomial of degree at most 2m − 1.
Example 5: NVE of Mathieu type. The standard Mathieu equation is

(3.35) ξ¨ = (c0 + c1 cos(ωt))ξ, ω = 0.
We cannot apply our method to compute the family of Hamiltonians correspond-
ing to this equation, because {c0 + c1 cos(ωt)} is not the general solution of any
differential polynomial with constant coefficients. But let us consider
d3 a da
(3.36) Q(a) = + ω2 ,
dt3 dt
the general solution of {Q(a) = 0} is
a(t) = c0 + c1 cos(ωt) + c2 sin(ωt).
Just notice that

c2
c1 cos(ωt) + c2 sin(ωt) = c21 + c22 cos ωt + arctan ,
c1
thus NVE (3.12), when a is a solution of (3.36), is reducible to Mathieu equation
(3.35) by a translation of time.
Using Proposition 3.7, we find the system of differential equations that deter-
mine the family of Hamiltonians,
d3 α dα d2 φ d2 α dφ dα
= 0, +3 − ω2 = 0.
dx31 2
dx1 dx1 dx1 dx1 dx1
The general solution of the first equation is
α = λ0 + λ1 x1 + λ2 x21 .
dφ
substituting it in the second equation, and writing y = dx1 , we obtain a non–
homogeneous linear differential equation for y,
dy 6λ2 y
(3.37) + = ω2 .
dx1 λ1 + 2λ2 x1
167
27
We must distinguish two cases depending on the parameter. If λ2 = 0, then we just

integrate the equation by trivial quadratures, obtaining
ω 2 x21
φ = µ0 + µ1 x1 +
2
and then,
y12 + y22 ω 2 x21
(3.38) H= + µ0 + µ1 x1 + − λ0 x22 − λ1 x1 x22 + β(x1 , x2 )x32 ,
2 2
If λ2 = 0, we can then reduce the equation to separable using
6λ2 y
u= ,
λ1 + 2λ2 x1
obtaining
3du 6λ2 dx 3ω 2 3µ1
= , u= + ,
3ω 2 − 4u λ1 + 2λ2 x1 4 4(λ1 + 2λ2 x1 )4
and then
1 2 2 µ1
y= ω λ1 + 2ω λ2 x1 + ,
8λ2 (λ1 + 2λ2 x1 )3
and finally we integrate it to obtain φ,

µ1 1 ω 2 λ1 x1 ω 2 x21
φ = ydx1 = µ0 − + + ,
32λ22 (λ1 + 2λ2 x1 )2 8λ2 8
scaling the parameters adequately we write down the general formula for the Hamil-
tonian,
y 2 + y22 µ1 λ1 ω 2 x1 ω 2 x21
H= 1 + µ0 + + + +
2 (λ1 + 2λ2 x1 )2 8λ2 8
(3.39) −λ0 x22 − λ1 x1 x22 − λ2 x21 x22 + β(x1 , x2 )x32 .
3.3.2. Non-integrability. One of the main results in [3] is the following.
Theorem 3.10 ([3]). The Galois group of the equation,
ξ¨ = Q(x)ξ,
with Q(x) a non-constant polynomial of degree k, with coefficient field K = C(x),
is a connected non-commutative group. In fact, the Galois group falls in either one
of the following cases:
(1) Case 4 of Proposition 2.5 for k = 2n even and provided some concrete
algebraic relations in the coefficients of Q are satisfied.
(2) SL(2, C) (case 7 of Proposition 2.5).
The proof of this theorem is given by means of a method of completion of
squares for the polynomial Q and by using the Kovacic algorithm (Appendix A,
Section A.1).
From Theorem 3.10 and Theorem 2.8 it one obtains the following non-integrability
result.
Theorem 3.11 ([3]). The Hamiltonian systems (3.31) with λ3 = 0, (3.33) with
n ≥ 1, (3.38) with λ1 = 0 and (3.39) with (λ1 , λ2 ) = (0, 0) are not integrable by
means of rational first integrals.
For further details see the original paper [3].
170
168
3.4. Celestial Mechanical Problems . Here we survey some recent results

about the non integrability of some Celestial Mechanical problems.
3.4.1. Some N -Body Problems. Using the Ziglin approach along the triangular
parabolic solution of Lagrange, A. V. Tsygvintsev ([131, 132, 133, 134, 135,
136]) proved the meromorphic non-integrability of the Three-Body Problem and
ultimately settled the non-existence of a single meromorphic first integral; he es-
tablished both things for all except three special cases. On the other hand, using
Theorem 2.7 also along the triangular parabolic solution of Lagrange and by means
of a result about the non-commutativity of the identity component of the Galois
group in presence of logarithmic terms for completely reducible variational equa-
tions (Appendix A, Section A.3), Boucher and Weil ([16, 17, 19]) also proved the
meromorphic non-integrability of the Three-Body Problem. It is finally worth not-
ing that Ziglin ([151]) managed to settle strong conditions on the integrability of
the Three-Body Problem and the equal-mass N -Body Problem.
In the already cited joint work of the first author with Simon ([103]) we are
reobtaining in simpler ways, strengthening and generalizing the results mentioned
in the previous paragraph. The proof is based on Theorems 3.3 and 3.1, i.e., we
use homothetical solutions. We follow [103].
The Hamiltonian of the general N -Body Problem in dimension d is
1 T −1
(3.40) HN,d (x, y) := y M y + UN,d (x) ,
2
defining
M = diag (m1 , . . . , m1 , · · · , mN , . . . , mN ) ∈ M at (N d, R) ,
and assembling the coordinates of our phase space among the N d-dimensional vec-
tors
x (t) = (xi (t))i=1,...,N , y (t) = (yi (t))i=1,...,N := (mi ẋi (t))i=1,...,N
of positions and momenta, respectively. The Newtonian gravitational potential is
mi mk
UN,d (x) := − .
xi − xk
1≤i<k≤N
The solutions of the equation

(3.41) UN d
(x) = λM x,
where λ > 0, are called central configurations. If the bodies are released with zero
initial velocity, with initial conditions at a (real) solution x of (3.41), this defines a
homothetical solution of the N -Body Problem.
We remark that we can normalize λ to one. Indeed, the −2 -homogeneity

of UN,d assures us UN,d (λα x) = λ−2α UN,d

(x); thus, assuming UN,d
(x) = λM x,

defining x̃ = λx and asking for UN,d (x̃) = M x̃ to hold, we obtain α = −1.
The problem of computing central configurations (i.e., solving the system of
algebraic equations (3.41), where we can assume λ = 1) is an old difficult classical
problem in Celestial Mechanics. In fact, it is only solved in complete generality for
N = 3, thanks to Euler (collinear central configurations: the masses are on a line)
and Lagrange (triangular central configurations: the masses are at the vertexes of
an equilateral triangle). For more information on central configurations see the
article [90].
169
29
From symmetry considerations, it is clear that whenever the masses are equal,
regular N -polygons with the masses at the vertexes give rise to homothetical so-
lutions, i.e., if the masses start with zero velocity from a such configuration, they
remain at a regular polygon. These are central configurations ([110]).
The connection of our work with homothetical solutions, and hence with central
configurations, is due to the following. For the N -Body Problem the real homo-
thetical points in Section 3.1 are central configurations and the particular integral
curves considered there are homothetical solutions. Indeed, a symplectic change x =
M −1/2 q, y = M 1/2 p renders HN,d a classical Hamiltonian HN,d = 12 p2 + VN,d (q)
with a potential which is homogeneous
of degree −1, VN,d = UN,d (M −1/2 q). Since

M 1/2 VN,d (q) = UN,d M −1/2 q and thus UN,d
(x) = M x (for x = M −1/2 q) is
equivalent to

VN,d (q) = M −1/2 M M −1/2 q = q.
Thus, we can consider the homothetical points in an N -Body Problem in Celestial
Mechanics as complex central configurations and the associated particular solution
Γ considered in Section 3.1 as an homothetical solution; this justify our terminology.
In virtue of Theorem 3.1, performing the following two steps would prove HN,d
not meromorphically integrable:
Step I either explicitly finding or proving the existence of an adequate constant
vector c ∈ C2N such that

(3.42) VN,d (c) = c;

Assume VN,d (c) is diagonalizable.

Step II proving that at least one of the eigenvalues of VN,d (c) does not belong to
the set given by items 1 and 18 in Table (3.3), which happens to be a set
of integers:

p (p − 3) (p + 2) (p − 1)
(3.43) S := − :p∈Z = − : p ∈ Z ⊂ Z,
2 2
whose symmetry allows for the assumption p > 1; the size of the consec-
utive gaps in this discrete set is strictly increasing, as is seen in its first
elements: {1, 0, −2, −5, −9, −14, −20, −27, −35, . . .}.
In virtue of Theorem 3.3, isolating an adequate set of eigenvalues and perform-
ing the following third step would be enough to set a very precise upper bound on
the amount of additional meromorphic integrals:
Step III proving that, except for a set S̃ of notable eigenvalues corresponding to

the set of classical first integrals, there is no other eigenvalue of VN,d (c)
in S.
And in virtue of either Theorem 3.2 or Theorem 3.3, the following fourth step
would be enough to discard the existence of even a single additional meromorphic
integral:
Step IV performing
III and proving that, except for said notable set S̃,
Step

Spec VN,d (c) \ S̃ consists exclusively of eigenvalues not satisfying rela-
tion (3.4) pairwise.
We are performing steps I–IV for the Three Body Problem with arbitrary
masses, steps I–III for the N-Body Problem with equal masses, N = 4, 5, 6, as
well as steps I and II for the N-Body Problem with equal masses with N ≥ 3. In
172
170
all cases we consider the planar case d = 2, although the proof for N = 3 can be
very easily established regardless of the dimension d ≥ 2 (see [103] for details).
a) Three Body Problem. Step I is computing a solution c of (3.42) for N =
3. Let us define m = m1 + m2 + m3 (which may be always set to 1 by the
reader if even simpler calculations are sought all through this section) and D =
m1 m2 + m2 m3 + m3 m1 , and consider vectors of the form c = m−2/3 M 1/2 ĉ, where
M = (mi Idd )i=1,...,N and
⎛ ⎞
a 2 m2 + a 3 m3
⎜ b2 m2 + b3 m3 ⎟
⎜ ⎟
⎜ a3 m3 − a2 (m1 + m3 ) ⎟
(3.44) ⎜
ĉ = ⎜ ⎟
⎟
⎜ b3 m3 − b2 (m1 + m3 ) ⎟
⎝ a2 m2 − a3 (m1 + m2 ) ⎠
b2 m2 − b3 (m1 + m2 )
and a2 , a3 , b2 , b3 are solutions to
3/2 3/2 3/2
a22 + b22 = a23 + b23 = (a2 − a3 )2 + (b2 − b3 )2 = 1.
An example of such a vector ĉ is

⎛ ⎞
(m2 + 2m3 ) α
⎜ m2 β ⎟
⎜ ⎟
⎜ − (m1 − m3 ) α ⎟
(3.45) ĉ = ⎜
⎜
⎟,
⎟
⎜ − (m1 + m3 ) β ⎟
⎝ − (2m1 + m2 ) α ⎠
m2 β
where α2 +β 2 = 1 and α3 = 1/8. The possible choices of α and β add up to two such
vectors as (3.45), and thus two solutions c = m−2/3 M 1/2 ĉ and √
c∗ = m−2/3 M 1/2 ĉ∗
∗ −1+i 3
for (3.42): those corresponding to α = 1/2 and α = 4 , respectively; where
square roots are taken in their principal determination. A simple, if tedious com-
putation proves c and c∗ solutions to (3.42), indeed. In fact, c yields an explicit
parametrization for the (homothetical) Lagrange triangular solution where the three
masses start at rest on the vertex of an equilateral triangle.
The rest of the proof is based on performing both Steps II and III at a time.
The eigenvalues of V3 (c) are {−2, 0, 0, 1, λ+ , λ− }, where

1 3 m21 + m22 + m23 − m1 m2 − m1 m3 − m2 m3
λ± := − ± .
2 2 (m1 + m2 + m3 )
It is not difficult to prove that the eigenvalues −2, 0, 0, 1 correspond to the classical
first integrals: the energy, the two components of the linear momentum and the
angular momentum (see [103]). Then as said in Theorem 3.3, one of the necessary
conditions for the existence of a single additional
meromorphic integral
for XH3
implies either λ∗+ ∈ S or λ∗− ∈ S, where S = − 12 p (p − 3) : p > 1 , which means
√
(defining R := m2 − 3D) that ±3R ∈ p2 − 3p − 1 m : p > 1 and therefore
(3.46)
−27 (m1 m2 + m1 m3 + m2 m3 ) ∈ m2 (p − 1) (p − 2) (p − 4) (p + 1) : p > 1 .
This is impossible if p ∈ {2, 4} or p > 4, since it would have a strictly negative
number equaling a non-negative one. For p = 3 (3.46) becomes 8m2 = 27D, that
171
31
is,
m1 m 2 + m1 m 3 + m2 m 3 8
(3.47) = .
(m1 + m2 + m3 )2 27
√
3√ A
The eigenvalues of V3 (c∗ ) are −2, 0, 0, 1, λ∗+ , λ∗− , where λ∗± = − 12 ± 2 2m
,
and
√
A = 2m21 + 2m22 + 2m23 − 5m1 m2 − 5m2 m3 + 7m1 m3 − i 3(m1 m2 + m2 m3 − 5m1 m3 ).
Again, the thesis in Theorem 3.3 amounts to either λ∗+ ∈ S or λ∗− ∈ S, which here
√ √
becomes ±3 A = p2 − 3p − 1 2m, and thus

2
A − 2m ∈ 2
(p − 1) (p − 2) (p − 4) (p + 1) m : p > 1 ;
2
9
a necessary condition for this to hold with real masses is the vanishing of the
imaginary term in A,
√
(3.48) −i 3 (m1 m2 + m2 m3 − 5m1 m3 ) = 0,
implying m1 m2 + m2 m3 = 5m1 m3 . Thus,
(3.49) −378m1 m3 = 2 (p − 1) (p − 2) (p − 4) (p + 1) m2 ,
for some p > 1. We discard p = 2, 4 in (3.49) assuming the strict positiveness of
m1 and m3 . The only integer p > 1 for which the right side can be negative is 3,
2
implying −378m1 m3 = −16 (m1 + m2 + m3 ) . These two constraints arising from
(3.48) and (3.49),
189
(3.50) 5m1 m3 = m1 m2 + m2 m3 , m1 m3 = (m1 + m2 + m3 )2 ,
8
cannot hold at the same time as condition (3.47). Indeed, the former two sub-
1 m3 +m1 m3 )
stituted into the latter would yield (5m189 = 278
, i.e. 16 8
63 = 27 which is
8 m1 m3
obviously absurd. Thus, either (3.47) holds or both equations in (3.50) hold.
Let us now prove that V3 does not satisfy the √remaining thesis in said Theorem.
The difference in (3.4), E (λi , λj ) = 9 − 8λj − 9 − 8λi /2, will be studied both
for Spec (V3 (c∗ )) and Spec (V3 (c)). Let
1/2 −1
a := m21 + m22 + m23 − m1 m2 − m1 m3 − m2 m3 (m1 + m2 + m3 ) ≥ 0.
The only case worth considering for the real eigenvalues is
√ √
13 + 12a − 13 − 12a
E (λ+ , λ− ) = ,
2
13
which is real only if a ∈ 0, 12 . In this interval, moreover, the only possible integer
√
values of E (λ+ , λ− ) are 0, 1, 2. Note that a = 1 − 3Q, where Q = D/m2 =
−2 √
(m1 m2 + m1 m3 + m2 m3 ) (m1 + m2 + m3 ) . The solutions to 1 − 3Q = n for
n = 0, 1, 2 are, respectively, Q = 1/3, 0, −1, among which the only possible value
for Q is 1/3. Hence, E (λ+ , λ− ) can only be real if a = 0,
√
i.e. Q = 1/3.
∗
Now consider the complex eigenvalues λ∗± = − 12 ± 3 2a of V3 (c∗ ). Since
⎛ ⎞
√ 1 + 12 a∗ − 1 − 12 a∗
E λ∗+ , λ∗− = 13 ⎝ ⎠,
13 13
2
174
172
it is enough to prove that (a∗ )2 is always never real when Q = 1/3. Indeed, if z =
√ √ √ √ 2
z1 +z2 i with z1 z2 = 0, then 1 + z− 1 − z is always complex: 1+z− 1−z =
√ √
2 − 2 1 − z 2 and since z 2 is non-real, so is 2 − 2 1 − z 2 .
In order to prove a∗ , (a∗ ) ∈ R \ C, we will see that the imaginary term inside
2
the square root, −5m1 m3 + m2 m1 + m2 m3 , is always nonzero if Q = 13 . Indeed,

5m1 m3 +m1 m3
otherwise (m +m +m )2
= 13 , i.e. 16m1 m3 − m21 − 2m2 m1 − m22 − 2m2 m3 − m23 = 0;
1 2 3
5m1 m3
from 5m1 m3 = m2 m1 + m2 m3 , we also deduce m2 = m1 +m3 and therefore
4m31 m3 − 15m21 m23 + 4m1 m33 − m41 − m43
16m1 m3 −m21 −2m2 m1 −m22 −2m2 m3 −m23 = = 0,
(m1 + m3 )2
and the only values of m3 allowing this are
√ √
(2 + 3i) ± (1 + 2i) 3 (2 − 3i) ± (1 − 2i) 3
m1 m1 ,
2 2
which are obviously not positive real numbers. The lack of an additional mero-
morphic first integral for arbitrary m1 , m2 , m3 > 0 is thus proven in the planar
case.
Hence, we have proven.
Theorem 3.12 ([103]). For the planar Three Body Problem, there is no addi-
tional meromorphic first integral with arbitrary positive masses which is independent
with the classical first integrals.
It is worth noting that the two cases forcing us to resort to a second solution
to (3.42) are precisely two of the three cases exceptional to Tsygvintsev’s proof
([136]):

D 1 23 2
(3.51) ∈ , , .
m2 3 33 32
b) N -Body Problem with equal masses. Here we consider the integrability

problem of N equal masses in a plane. Defining q = (q1 , . . . , qN ) (qi = (q2i−1 , q2i )),
we have
n
∂VN,2 √ 3/2 −3
(3.52) = mk (mi mk ) Di,k Di,k , i = 1, . . . , N,
∂qi
k=1,k=i
√ √
where Di,j = (d2i−1,2j−1 , d2i,2j )T :=mj qi − mi qj for each i, j = 1, . . . , N , and

we obtain the block expression for the Hessian matrix: VN,2 (q) = Ũi,j ,
i,j=1,...,N
defining
√
− mi mj Ui,j ,
i = j,
(3.53) Ũi,j :=
k=i mk Ui,k , i=j
where

02×2 , i = j,
(3.54) Ui,j = Uj,i = 3/2 −5/2
(mi mj ) d22i−1,2j−1 + d22i,2j Si,j , i < j,
being 02×2 the zero square matrix of dimension two and
2
d2i,2j − 2d22i−1,2j−1 −3d2i−1,2j−1 d2i,2j
(3.55) Si,j = Sj,i := , i = j.
−3d2i−1,2j−1 d2i,2j d22i−1,2j−1 − 2d22i,2j
173
33
For simplicity, we will denote VN := VN,2 from now on.

Since every choice of mass units amounts to a symplectic change in the extended
phase space, we may set m1 = · · · = mN = 1. Expressions (3.52) and (3.53) may be
found explicitly in terms of trigonometric functions if we choose the regular central
polygonal configuration, where each of the masses are at the vertex of a regular
polygon, as a solution to (3.42). Define
πk πk
sk := sin , ck := cos , k ∈ N,
N N
2πi
and ζ = e N = c2 + is2 .
1/3
Lemma 3.13. Vector cP = (c1 , . . . , cN ) defined by cj = βN (c2j , s2j ), where
N −1
βN = 14 k=1 csc πk
N , is a solution for VN (q) = q.

Proof. Indeed, assume cj = A cos 2πj 2πj
N , sin N for some A > 0. We have
⎛ 2πj
⎞
N −1 cos N
∂VN 1 ⎝ π
k=1 sin N k ⎠
∂qj
(cP ) =
4A2 N −1 sin 2πj N
,
π k=1 sin N k
due to the fact that

N
N
−1
ζj − ζk 1 − (c2k + is2k )
3 = ζj 3 ,
k=1,k=j |ζ j − ζk| k=1 |1 − ζ k |
and, since the imaginary part of this sum satisfies:
N
−1 N
−1 N −1
s2k 2sk ck 1 ck
= = = 0,
k 3 8c3k s2
k=1 |1 − ζ |
4
k=1 k=1 k
N −1 N −1
we finally obtain ζ j k=1 1−(c 2k +is2k )
|1−ζ k |3
= 14 ζ j k=1 s−1
k . Now V (cP ) = cP if and
N −1 1 1/3
only if k=1 4A2 sk = A. The latter holds for A = βN .
Let us see how this specific vector simplifies VN . Keeping expression (3.53) in

consideration we have d2i−1,2j−1 + id2i,2j = βN ζ i − ζ j which implies
1/3
2 3c
1/3 2(i+j) − 1 3s2(i+j)
Si,j = 2 βN si−j ,
3s2(i+j) −3c2(i+j) − 1
for each 1 ≤ i, j ≤ N , and thus
Ui,i = 02×2 , i = 1, . . . , N,
−5
1/3
Ui,j = Uj,i = 2βN si−j Si,j
−3
|si−j | 3c2(i+j) − 1 3s2(i+j)
= , i = j,
16βN 3s2(i+j) −3c2(i+j) − 1
from which defining
|si−j |−3 3c
2(i+j) − 1 3s2(i+j)
Ũi,i = ,
16βN 3s2(i+j) −3c2(i+j) − 1
j=i

|si−j |−3 1 − 3c2(i+j) −3s2(i+j)
Ũi,j = , i= j,
16βN −3s2(i+j) 3c2(i+j) + 1
176
174

we have VN (cP ) = Ũi,j .
i,j=1,...,N
Lemma 3.14. The trace for VN (cP ) is equal to −(N/8) (αN /βN ), where αN =
N −1 3 πk

k=1 csc N and βN is defined as in Lemma 3.13.
Proof. In virtue of the above simplifications for (3.53), tr (VN (cP )) is equal
to
2 2k
µN := − ζ 1 − ζ 2k2 −3 .
βN
1≤k1 <k2 ≤N
N −1 −3
We have − µ4N k=1 csc πk N = 1≤k1 <k2 ≤N 2ζ 2k1 − ζ 2k2 ; on the other hand,
the symmetry of a regular polygon assures
N
−1
−3 −3
2 |2sk2 −k1 | =N (2sk ) ;
1≤k1 <k2 ≤N k=1
N −1 πk
N −1 πk

thus, 2µN k=1 csc N = −N k=1 csc3 N .
Case 1: N = 3, 4, 5, 6.
We can afford obtain a result stronger than non-integrability for these values
without using Lemma 3.14. We just have to prove the following
Lemma 3.15. VN (cP ), N = 3, 4, 5, 6, has only four eigenvalues in S: λ1 =

−2, λ2 = λ3 = 0, λ4 = 1.Furthermore, the sets of equivalence classes given by
relation E (λi , λj ) ∈ Z in (3.4) with cardinality greater than one are (assuming
j > 4):
(1) a double eigenvalue for N = 3, 4;
(2) three double eigenvalues for N = 5, 6.
Proof. The eigenvalues of V3 (cP ) are√λ1 , λ2 , λ3 , λ4√and λ5,6 = −1/2. Those
2(5−3 2) 2( 2−4) √
of V4 (cP ) are λ1 , λ2 , λ3 , λ4 and λ5 = 7 , λ6,7 = 7 , λ8 = 6 2−17
7 . The
corresponding relations are
√ √
2 5−3 2 2 2−4 1 √ 1 √
E , = − −119 + 336 2 + 889 − 112 2,
7 7 14 14
√ √
6 2 − 17 2 2 − 4 1 √ 1 √
E , = − 1393 − 336 2 + 889 − 112 2,
7 7 14 14
√ √
6 2 − 17 2 5 − 3 2 1 √ 1 √
E , = − 1393 − 336 2 + −119 + 336 2.
7 7 14 14
V5 (cP ) has six different non-trivial double eigenvalues:

√ √ √
5 − 5 ± 518 − 222 5 5−4
λ5,6,7,8 = , λ9,10 = .
4 2
175
35
Relations are
√ √ √ √ √ √ √ √
19 − 2 5 + 6 37 − 2 37 5 − 19 − 2 5 − 6 37 + 2 37 5
E (λ5,6 , λ7,8 ) = ,
2
√ √ √ √
25 − 4 5 − 19 − 2 5 − 6 37 + 2 185
E (λ5,6 , λ9,10 ) = ,
2
√ √ √ √
25 − 4 5 − 19 − 2 5 + 6 37 − 2 185
E (λ7,8 , λ9,10 ) = .
2
The eight non-trivial eigenvalues for V6 (cP ) are

√ √ √
4 29 3 − 94 34 3 − 133465 − 59584 3 − 157
λ5 = , λ6,7 = ,
59 118
√ √ √
2 7 3 − 41 34 3 + 133465 − 59584 3 − 157
λ8,9 = , λ10,11 = ,
59 √ 118
4 53 − 22 3
λ12 = .
59
The relations are
√
s1 + 236s2 − s3 s4 − s3
E (λ5 , λ6,7 ) = , E (λ5 , λ8,9 ) = ,
√ 118 118
s1 − 236s2 − s3 s5 − s3
E (λ5 , λ10,11 ) = , E (λ5 , λ12 ) = ,
√ 118 118
s1 − 236s2 − s4 s5 − s4
E (λ8,9 , λ10,11 ) = , E (λ8.9 , λ12 ) = ,
√ 118 118√
s4 − s1 + 236s2 s5 − s1 + 236s2
E (λ6,7 , λ8,9 ) = , E (λ6,7 , λ12 ) = ,
√ 118 118
s5 − s1 − 236s2
E (λ10,11 , λ12 ) = ,
√ 118 √
s1 − 236s2 − s1 + 236s2
E (λ6,7 , λ10,11 ) = ,
118
√ √ √
with s1 = 68381 − 8024 3, s2 = 133465 − 59584 3, s3 = 208801 − 54752 3,
√ √
s4 = 70033 − 6608 3, s5 = −68735 + 41536 3.
Let us now determine an upper bound for the amount of meromorphic first
integrals for the equal–mass Problem.
(1) For N = 3, we have Λ1 = {λ5,6 } = {1/2}. Then K = 2, r = 1, which
by Theorem 3.3 in an absence of further information would allow the
existence of an additional meromorphic first integral. That possibility,
however, is ruled out by Theorem 3.12.
(2) For N = 4, we have Λ1 = {λ6,7 } and two simple eigenvalues: {λ5 } and
{λ8 }: Thus K = 2, r = 1 and there can be at most one additional
meromorphic first integral.
(3) For N = 5, we have Λ1 = {λ5,6 }, Λ2 = {λ7,8 }, Λ3 = {λ9,10 }. Then K = 6
and r = 3. Hence, there may be at most three additional meromorphic
integrals.
(4) For N = 6, we have Λ1 = {λ6,7 }, Λ2 = {λ8,9 }, Λ3 = {λ10,11 }, and two
simple eigenvalues {λ5 } and {λ12 }. As for N = 5, we have at most three
additional meromorphic integrals.
178
176
Case 2: N = 7, 8, 9.
Proceeding from Lemma 3.14, it is straightforward to see the traces for VN (c)
for these three values of N are non-integers since
√ √
413 + 56 7 cos 13 arctan 3 3
µ7 = − √ ∈ (−12, −11) ,
2 cos 16 arctan 3133
√ √
4 −2633 + 766 2 + 4 118010 − 68287 2
µ8 = ∈ (−17, −16) ,
√
241
9 8 9 3 + csc3 π9 + csc3 2π 3 4π
9 + csc 9
µ9 = − √ ∈ (−22, −21) .
2 2 3 + csc π + csc 2π + csc 4π
3 9 9 9
Case 3: N ≥ 10
We will prove VN (cP ) has at least an eigenvalue greater than 1. We know the
following holds ([1]),
k−1
1 1 (−1) 2 22k−1 − 1 B2k x2k−1
(3.56) csc x = + f (x) := + ,
x x (2k)!
k≥1
f being analytical for |x| < π (which obviously holds if x = πjN , j = 1, . . . , N − 1)

and Bk , k ≥ 1, being the Bernoulli numbers ([1, section 23]).
N −1 jπ
Lemma 3.16. For each N ≥ 10, SN := 2 j=1 csc2 jπ N − 5 csc N > 0.
Proof. Recall the Euler-MacLaurin summation formula ([128, §3.3]): for any
f ∈ C 2s+2 ([a, b]) and n ∈ N, and defining h = b−a
n , the following holds,
n b s
f f (a) + f (b) 2r−1 f (2r−1) (b) − f (2r−1) (a)
a
f (a + jh) = + + h B2r + Rs ,
j=0
h 2 r=1
(2r)!
B2s+2 (2s+2)
where Rs = nh2s+2 (2s+2)! f (α) for some α ∈ (a, a + nh). Substituting in

a = h = π/N , n = N − 2, b = a + hn = π(NN−1) , f (x) = 2 csc2 x − 5 csc x and
s = 2, we obtain
2N π
b
a
f (x) dx π π
= cot csc + 9 ln tan ,
h π N N 2N
f (a) + f (b) π π
= 2 csc2 − 5 csc ,
2 N N
π π π
f (b) − f (a) π cot N csc N 3 csc2 N −5
hB2 = ,
2 3N
π π
f (b) − f (a)
π 3 csc6 N 742 cos N + 213 cos 3π N + 5 cos N
5π
h 3 B4 = −
4! 2880N 3
π π
π (742 + 213 + 5) csc6 N
3
π 3 csc6 N
> − = − ,
2880N 3 3N 3
and
csc9 (α) (N − 2) π 6 P (α)
R2 (α) = ,
1935360N 6
177
37
where P (x) := 1110231 + 1256972 cos 2x + 206756 cos 4x + 6516 cos 6x + 5 cos 8x;.
In previous formulae, we have used B2 = 1/6, B4 = −1/30, B6 = 1/42 and several
trigonometric identities in order to express the different terms in a suitable way for
what follows.
Introducing variable w = cos 2x, we may write the function defined by the first
three terms in P (x) as
P! (w) := 903475 + 1256972w + 413512w2 .
Then, for each w ∈ [−1, 1], one has P! (w) > 0; hence, for x ∈ (0, π) we obtain
P (x) ≥ P! (−1) − 6516 − 5 > 0 and therefore R2 (α) > 0, which leads to the
following:
b
f (a) + f (b) 2r−1
2
a
f f (2r−1) (b) − f (2r−1) (a)
SN = + + h B2r + R2 (α)
h 2 r=1
(2r)!
b
f (x) dx f (a) + f (b) 2r−1
2
a f (2r−1) (b) − f (2r−1) (a)
> + + h B2r
h 2 r=1
(2r)!
π π π

2N cot N csc N + 9 ln tan 2N π π
> + 2 csc2 − 5 csc
π N N
π π π π
π cot N csc N 3 csc2 N −5 π 3 csc6 N
+ − .
3N 3N 3
There is a number of possible ways of proving this latter lower bound strictly
π
positive. For instance, since, for N ≥ 10, cot N > 3, we have
2N π π π π π
SN > cot csc + 9 ln tan + 2 csc2 − 5 csc
π N N 2N N N
π π π π 3 csc6 π
N
+ csc 3 csc2 −5 −
N N N 3N 3
=: σN .
π

The first term in that sum is exactly 2N π F tan 2N , where
z −2 − z 2
F : (0, ∞) → R, F (z) := + 9 ln z,
4
√ π
√
is strictly decreasing in 0, 5 − 2 . Since tan 2N < 5 − 2 for all N ≥ 10, we have
π π 20
F tan ≥ F tan >− ,
2N 20 3
and thus,

2N 20 π π π π π π 3 csc6 π
N
σN > − + 2 csc2 − 5 csc + csc 3 csc2 − 5 −
π 3 N N N N N 3N 3
csc Nπ π
> G N csc ,
3N 3 N
where GN (x) := −π 3 x5 + 3N 2 (2N + 3π) x2 − N 2 (55N + 15π) and we have used
π
csc (x) > x1 for all x ∈ (0, π) (see (3.56)) and thus − 40N
3π > − 40
3 csc N for all
N ≥ 2. It is immediate that GN (x) > 0 if
1/3
N 12 + 18 Nπ2 N4
x ∈ 0, ⊃ 0, .
π 5 π 3
180
178

For all N ≥ 3, the
latter interval contains N π
π , csc N , thus allowing us to lower-
π
bound GN csc N by

N N5 9π 55π 2 15π 3
GN = 2 −1 + 6 + − − > 0, N ≥ 10.
π π N N2 N4
csc( π ) π

In this way we obtain SN > σN > 3NN3 G csc N > 0, N ≥ 10.
Lemma 3.17. For N ≥ 10, VN (cP ) has at least one eigenvalue greater than 1.
Proof. Indeed, let A = (ai,j )i,j=1,...,2N = VN (cP ). The Rayleigh quotient for
T
vector v = e2N,2N −1 = (0, 0, · · · , 0, 1, 0) is
N −1 π
π

v T Av vT ŨN,N vN j=1 csc3 j N 3 cos 2j N −1
= N T = a2N −1,2N −1 = −1 ,
T
v v vN vN 4 N π
j=1 csc j N
and it will be strictly greater than 1 if and only if
N
−1 N−1 N
−1
2jπ 3 jπ jπ 2 jπ jπ
3 cos − 1 csc −4 csc = 2 csc − 5 csc > 0,
j=1
N N j=1
N j=1
N N
which we already know holds for N ≥ 10 by Lemma 3.16. Elementary Linear
Algebra then yields the existence of at least one eigenvalue λ̃ > 1 for VN (cP ).
Since max S = 1 < λ̃, λ̃ ∈
/ S, we have proved that the planar equal masses
N -Body Problem with N ≥ 3 is not integrable by meromorphic first integrals.
Summarizing, we have proven:
Theorem 3.18 ([103]). For the planar equal masses N - Body Problem the
following assertions hold:
1. The number of additional meromorphic first integrals is no greater than:
a) one if N = 4;
b) three if N = 5, 6.
In particular, the Problem is not meromorphically integrable in the sense
of Liouville for all three values of N .
2. For N ≥ 3 problem is not meromorphically integrable in the sense of
Liouville. For N = 3, 4, 5, 6, there is no additional meromorphic first
integral which is independent with the classical first integrals.
For more details, see the original paper [103].
We must observe that Hamiltonian HN,d is not meromorphic. However, any
first integral of XHN,d (e.g. HN,d itself), when restricted to a domain of each
determination of HN,d , is meromorphic and thus amenable to the whole theory
explained so far; see, for instance, [73, pp. 156-157] for more details as applied to
a different homogeneous potential.
3.4.2. Hill’s problem. In a joint work with Simó and Simon the first author
proved the non–integrability of Hill’s problem [100]. We follow this reference hence-
forth. As the proof is technical and strongly based on the Galoisian correspondence,
normality of the Picard-Vessiot extensions (see Section 2.2) and algebraic groups,
we only review in detail the computational steps in the proof, in agreement with
our approach, where we are interested mainly on algorithms and methodology.
Hill’s problem, usually called lunar as an homage to its earliest motivation, or
planar in order to distinguish it from its own extension to R3 , is a model originally
179
39
based on the Moon’s motion under the joint influence of Earth and Sun ([45]). We
can think of the Hill’s problem as a limiting case of the Restricted Three Body
Problem for a negligible mass of the Earth and when the distance of the Sun tends
to infinity. After some manipulations using the Levi-Civita regularization it is
possible to write the Hamiltonian of this problem as a polynomial of degree six
([124]):
(3.57) H(Q, P ) = H2 + H4 + H6,

a sum of homogeneous polynomials of degrees 2, 4 and 6, respectively:
1 2 1
H2 = (P + P22 ) + (Q21 + Q22 ), H4 = −2(Q21 + Q22 )(P2 Q1 − P1 Q2 ),
2 1 2
H6 = −4Q2 (Q41 − 4Q21 Q22 + Q42 ),

where, as usual, Q = (Q1 , Q2 ) are the positions and P = (P1 , P2 ) are the momenta;
here we prefer to use this notation in agreement with ref. [124], rather than x and
y as in other parts of the text.
Now we recall that our method (up to first order, i.e., we only need here the first
order obstruction to integrability given by V E := V E1 ) is given by the following
steps:
(1). Find a particular integral curve.
(2). Write the V E.
(3). Check if the identity component of the differential Galois group of the V E
is commutative.
(1) Particular integral
curve.
We find a particular integral curve as follows.
1 1 i
The matrix A = √2 provides for a symplectic change of variables,
i 1

Q1 Q̄1 P1 T P̄1
=A , = A−1 ,
Q2 Q̄2 P2 P̄2
which in turn transforms Hamiltonian (3.57) into
H̄ = i(Q̄1 Q̄2 − P̄1 P̄2 ) − 4i(3Q̄41 − 2Q̄21 Q̄22 + 3Q̄42 )Q̄1 Q̄2 − 4Q̄1 Q̄2 (Q̄1 P̄1 − Q̄2 P̄2 ).
The corresponding differential system z̄ = XH̄ (z̄) now displays two invariant planes

π1 : Q̄2 = P̄1 = 0 , π2 : Q̄1 = P̄2 = 0 ,
in any of which all nontrivial information of that system reduces to a hyperelliptic
equation,
(3.58) φ = −φ + 12φ5 ,
which through multiplication by φ and subsequent integration becomes
(φ ) = −φ2 + 4φ6 + 2h.
2
(3.59)
Defining w = φ2 , z = 2φφ , we arrive to the system

(3.60) w = z, z = 4 −w + 8w3 + h ,
whose Hamiltonian (at level zero energy) is K(w, z) = 12 z 2 + 2w2 − 8w4 − 4hw.
182
180
The solution to system (3.60), or equivalently to equation (w ) = −4w2 +

2
4
16w + 8hw, is the inverse of an elliptic integral:
w(t)
t=± (−4y 2 + 16y 4 + 8hy)−1/2 dy + C1 , C1 ∈ C,
0
translation t → t − C1 being the next obvious step. It is a known fact (see, for
instance, [141]) that given a polynomial of degree four without repeated factors,
p4 (x) = a4 x4 +4a3 x3 +6a2 x2 +4a1 x+a0 , and defining constants (called invariants)
g2 = a4 a0 − 4a3 a1 + 3a22 , g3 = a0 a2 a4 + 2a1 a2 a3 − a32 − a4 a21 − a23 a0 ,
w(t)
the solution for t = a (p4 (x))−1/2 dx is the following:

p4 (a)℘ (t; g2 , g3 ) + 12 p4 (a) ℘(t; g2 , g3 ) − 24
1 1
p4 (a) + 24 p4 (a)p
4 (a)
w(t) = a + 1
2 (4)
,
2 ℘(t; g2 , g3 ) − 24 p4 (a) − 48 p4 (a)p4 (a)
1
where ℘(t; g2 , g3 ) is the Weierstrass elliptic function. In our specific case, this
becomes
w(t) = 6h/F (t), z(t) = −18h℘ (t; g2 , g3 )/F 2 (t),
where F (t) := 3℘(t; g2 , g3 ) + 1. In particular,

φ1 (t) = 6h/F (t), φ2 (t) = −φ1 (t),
are solutions
√ to original equation (3.58). Furthermore, a simple calculation proves
h∗ = 1/(6 3) to be a separatrix value in which φ21 (t) = φ22 (t) degenerates into
combinations of hyperbolic functions. We assume 0 < h < h∗ .
We are now proving that, for the above range of h, w(t) has two simple poles in
each period parallelogram, the sides of which will be denoted as 2ω1 , 2ω2 , as usual.
In virtue of [35, p. 96], expression 1/(℘(t) − ℘(t∗ )) (in our case, ℘(t∗ ) = −1/3) has
exactly two simple poles in t∗ , −t∗ (mod 2ω1 , 2ω2 ), with respective residues 1/℘ (t∗ )
and −1/℘ (t∗ ). Therefore, all double poles, if any, of 1/(℘(t) − ℘(t∗ )), expanding
around t = t∗ , are precisely those t∗ such that ℘ (t∗ ) = 0. We have
4 8
(℘ (t; g2 , g3 ))2 = 4(℘(t; g2 , g3 ))3 − g2 ℘(t; g2 , g3 ) − g3 = 4℘3 − ℘ − + 64h2 ,
3 27
and every pole (whether double or not) must satisfy ℘(t∗ ) = −1/3; X = −1/3 is
obviously not a root of 4X 3 − 4X/3 − 8/27 + 64h2 unless h = 0.
So, just we proved that the Hamiltonian system defined by (3.57) has a partic-
ular integral curve (depending on the energy level h) of the form
1
(3.61) (Q1 (t), Q2 (t), P1 (t), P2 (t)) = √ (φ(t), iφ(t), φ (t), iφ (t)) .
2
√ 2
For all 0 < h < 1/ 6 3 , φ (t) is an elliptic function with two simple poles in each
parallelogram period.
(2) Variational equation. We compute the variational equation along the above
integral curve. Reordering the vector of dependent canonical variables as (Q̄1 , P̄2 , Q̄2 , P̄1 )T
and restricting ourselves to the particular solution,
Q̄1 = φ, Q̄2 = 0, P̄1 = 0, P̄2 = iφ ,
181
41
the variational equations (VE ) along that solution are written as

(3.62)
⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
ξ¯ 0 −i −4w 0 ξ¯ ξ¯
⎜ η̄ ⎟ ⎜ i(60w2 − 1) −4iz 4w ⎟ ⎜ ⎟ ⎜ η̄ ⎟
⎜ ⎟=⎜ 0 ⎟ ⎜ η̄ ⎟ =: A1 B1 ⎜ ⎟
⎝ ξ ⎠ ⎝ 0 0 0 −i ⎠ ⎝ ξ ⎠ 0 A1 ⎝ ξ ⎠,
η 0 0 i(60w2 − 1) 0 η η
and their lower right block, the normal variational equations (NVE )

ξ 0 −i ξ
(3.63) = ,
η i(60w2 − 1) 0 η
that is,
(3.64) ξ (t) = (60w2 (t) − 1)ξ(t).
Next step is to obtain a fundamental matrix for (3.63). An obvious shortcut is to
take w as new independent variable and to define Ξ(w), H(w) such that ξ = Ξ ◦ w
and η = H ◦ w. We have

d2 Ξ w − 8w3 − h dΞ 60w2 − 1
(3.65) = 4 + Ξ,
dw2 wf (w, h) dw wf (w, h)
also expressible in matrix form
d
Ξ 1 0 −i Ξ
(3.66) dw
d = ,
dw H wf (w, h) i(60w 2
− 1) 0 H
where f = f (w, h) = 4(4w3 − w + 2h).
The VE, equation (3.62), is integrable; of course, we know from section 2 that
this is due to the solvability of the identity component of the Galois group, see
below.
Let us start from the block notation

P Q
(3.67) Ψ= ,
R S
P, Q, R, S being 2 × 2 matrices with their entries in some differential field to be
described below. We can assume Ψ(0) = Id4 , which, along with the triangular form
of (3.62), assures R ≡ 0. In particular, the matrix form of the NVE (3.63) can be
written as S = A1 S. Let us now proceed to integrate these normal equations. More
precisely, let us collect all necessary information about the fundamental matrix
ΦN (t) of (3.63) with initial condition ΦN (0) = Id2 .

Using well-known properties of ℘ and ℘ , it is easy to prove that Ξ1 (w) =
f (w, h) is a solution of (3.65), and therefore
−3/2
ξ1 (t) = Ξ1 (w(t)) = ℘ (t; g2 , g3 ) (3℘(t; g2 , g3 ) + 1)
is a solution of (3.64). A first solution of (3.63) is then

ξ1 (t) 16w3 (t) − 4w(t)
+ 8h
= C1 , C1 ∈ C.
η1 (t) −2i 12w2 (t) − 1 w(t)
We now recall d’Alembert’s method ([48, p. 122]) in order to obtain a second
solution of (3.64) independent of ξ1 . This solution is
t
(3.68) ξ2 (t) = ξ1 (t) {ξ1 (τ )}−2 dτ.
0
184
182
After recovering our former independent variable t through composition, we have

a fundamental matrix for the NVE, that is, the block S in (3.67),

ξ1 ξ2 ξ1 ξ2
ΦN (t) = = .
η1 η2 iξ1 iξ2
In particular, P (t) ≡ S(t) since they are both fundamental matrices for the same
initial value problem. We now compute the block Q in (3.67); the standing equa-
tions (in vector form) are

ξ¯ 0 −i ξ¯ −4w 0 ξ
(3.69) = + ,
η̄ i(60w2 − 1) 0 η̄ −4iz 4w η
where (ξ, η)T are the solutions to the NVE. Applying variation of constants to
(3.69) we obtain
t
(3.70) Q(t) = ΦN (t) V (τ )dτ,
0
where
−4w(t) 0
C(t) = , V (t) = Φ−1
N (t)C(t)ΦN (t).
−4iz(t) 4w(t)
In other words, the fundamental matrix of (3.62) has the form
t
ΦN (t) ΦN (t) 0 V (τ )dτ
(3.71) Ψ(t) = .
0 ΦN (t)
In view of (3.70), computing Ψ explicitly would now only take the computation
of four integrals. The path we are taking, however, is a different one, although we
are keeping in mind all of this notation and the final expression (3.71).
Our next aim is to prove only two specific properties of the fundamental matrix
Ψ of (3.62), namely the existence of first and second class elliptic integrals and
logarithmic terms in its coefficients.
Let K be the field of all elliptic functions of the complex plane (with some fixed
periods). We know a solution of (3.64),

ξ1 (t) = 4w3 (t) − w(t) + 2h,
and can obtain a second one using (3.68) and the chain rule. Let us define α1 , α2 , α3
as the values of w for which f (w, h) = 0, the functions
" "
w(α3 − α1 ) α3 (α1 − α2 )
β(w, h) := arcsin , k(h) := ,
α3 (w − α1 ) α2 (α1 − α3 )
(both attaining complex, nonzero values if h ∈ (0, h∗ ) and therefore w(t) = 0) and
let
β β
1 1
E(β|k) := (1 − k2 sin2 θ)− 2 dθ, F (β|k) := (1 − k2 sin2 θ) 2 dθ.
0 0
be the elliptic integrals of first and second class, respectively (see [35], [141]). We
then obtain a fundamental matrix for the NVE (3.66),

Ξ1 (w) Ξ2 (w)
ΦN (w) = =
H1 (w) H2 (w)
183
43

f (w, h) g1 {f1 E(β|k) + f2 F (β|k) + g2 }
√ d
,
2i w(−1 + 12w2 ) i dw (g1 {f1 E(β|k) + f2 F (β|k) + g2 })
for some f1 = f1 (h), f2 = f2 (h), g1 = g1 (w, h), g2 = g2 (w, h), the first three non-
vanishing if h ∈ (0, h∗ ), and the last two linked to w by algebraic equations. In
particular, this yields our fundamental matrix ΦN (t) = ΦN (w(t)) for (3.63).
The fundamental trait of E(β|k) and F (β|k) is that they are transcendental over
K. Indeed, nontrivial elliptic integrals of the first and second classes are not elliptic
functions (see [35, Theorem 6.5 and its proof]) and they stem from quadratures;
thus, as we said before, E(β|k) and F (β|k) cannot be expressed in terms of elliptic
functions under any relation of algebraic dependence.
Let us prove the existence of terms with nonzero residue in the diagonal of
matrix V (t). Since

ξ1 ξ2 ξ1 ξ2
ΦN (t) = =
η1 η2 iξ1 iξ2
is the fundamental matrix of a Hamiltonian linear system, it is symplectic. The
integrand in (3.70) becomes

−w(ξ2 ξ1 + ξ1 ξ2 ) + w ξ1 ξ2 −ξ2 (2ξ2 w−ξ2 w )
V (t) = 4 i =:
(2wξ1 −w ξ1 )ξ1 w(ξ1 ξ2 +ξ1 ξ2 )−w ξ1 ξ2

u(t) v1 (t)
4i .
v2 (t) −u(t)
For every h ∈ (0, h∗ ), we expand these four entries around a simple pole t∗ of w(t);
expressing only the first term in each power series, we have
w(t) = C0 (t − t∗ )−1 + O(1),

ξ1 (t) = 2C0 (t − t∗ )−3/2 + O (t − t∗ )−1/2 ,
3/2
C0
−3/2
ξ2 (t) = (t − t∗ )5/2 + O (t − t∗ )7/2 ,
8
for some C0 = C0 (h) ∈ C; therefore,
C0
u(t) = − (t − t∗ )−1 + O(1),
2
3
v1 (t) = − (t − t∗ )3 + O (t − t∗ )4 ,
32C02

v2 (t) = −8C04 (t − t∗ )−5 + O (t − t∗ )−4 .
Hence, and save for the only value of h forcing C0 = 0 (i.e. h = 0), we have a
nonzero residue in u (t), which results in the aforementioned logarithmic terms in
the diagonal of
t t t
0
u(τ )dτ 0 1
v (τ )dτ
V (τ )dτ = t t
.
0
0
v2 (τ )dτ − 0
u(τ )dτ
We have now obtained a class of functions which cannot be linked algebraically
to the former. Indeed, logarithms are special cases of elliptic integrals of the third
class, which are neither elliptic functions nor elliptic integrals of first or second class
186
184
(see [35, Theorem 6.5 and its proof] once more), and in this case the logarithms
have been obtained through a quadrature.
Summarizing, we have obtained the following information about the funda-
mental matrix of the variational equation. The variational equation along solution
(3.61) have a fundamental matrix of the form
t
ΦN (t) ΦN (t) 0 V (τ )dτ
Ψ(t) = ,
0 ΦN (t)
where
ξ1 (t) ξ2 (t)
ΦN (t) =
iξ1 (t) iξ2 (t)
is a fundamental matrix of the normal variational equation; furthermore, ξ2 is a
linear combination of elliptic functions and nontrivial elliptic integrals of first and
t
second classes, and 0 V (τ )dτ is a 2 × 2 matrix function containing logarithmic
terms in its diagonal.
Let us interpret our results in terms of field extensions. First of all, we note that
using coordinates (x, y) = (φ, φ ) all solutions of the equation (3.59) are included
in the hyperelliptic curve

Γh := (x, y) ∈ C2 : y 2 = −x2 + 4x6 + 2h .
The previous transformation w = x2 , z = 2xy induces a finite branched covering
Γh → Λh ,
where Λh is the elliptic curve defined by
Λh := {z 2 /2 + 2w2 − 8w4 − 4hw = 0}.
Keeping K (= M(Λh )) as the field of all elliptic functions, let us describe the
Picard-Vessiot extension over K for VE Γh in detail
(1) First of all, let us define the extension
K ⊂ K1 := K(ξ1 , ξ1 ),
based on the adjunction of the first solution ξ1 of (3.63) and its derivative,
which is an algebraic (in fact, quadratic) one.
(2) Second of all, adjoining the solution ξ2 from (3.68) to this new field, we
obtain the extension
K1 ⊂ L1 := K1 (ξ2 , ξ2 ) = K(ξ1 , ξ1 , ξ2 , ξ2 ),
which is transcendental, since it is nontrivial and defined exclusively by
an adjunction of quadratures.
(3) Third of all, adjoining the matrix integral from (3.70) to L1 , we have
t t t
L1 ⊂ L2 := L1 u, v1 , v2 ,
0 0 0
also given by quadratures, nontrivial, and thus transcendental.
So far, the Picard-Vessiot extension L2 | K of the (3.62) splits as follows
K ⊂ K 1 ⊂ L1 ⊂ L2 .
(3 ) The identity component of the Galois group G of (3.62) is not com-

mutative. This step is a bit more technical and we do not show the details here;
we only comment that it is based on the following facts:
185
45
(a). By definition, G is the Galois group of the Picard-Vessiot extension Gal(L2 /K).
(b). By the Galoisian correspondence G0 = Gal(L2 /K2 ), since K2 ⊂ L2 is tran-
scendental; see Section 2.2.
(c). A detailed analysis of the extension K2 ⊂ L2 , again using the Galoisian
correspondence, proves that in the representation of G0 given by its action on the
fundamental matrix of (3.62) this group is a non-commutative subgroup of the
unipotent group
⎧⎛ ⎞ ⎫
⎪
⎪ 1 µ −κ + µβ γ + µκ ⎪
⎪
⎨⎜ ⎟ ⎬
! 0 ⎜ 0 1 β κ ⎟ : µ ∈ C, κ ∈ S1 , β ∈ S2 , γ ∈ S3 ,
(3.72) G = ⎝ ⎠
⎪
⎪ 0 0 1 µ ⎪
⎪
⎩ ⎭
0 0 0 1
where a fundamental role is played by the fact that the extension K2 ⊂ L2 splits
in K2 ⊂ L1 and L1 ⊂ L2 , the former given by non-trivial elliptic integrals and the
latter with a logarithmic term. See the original paper [100] for the details. As a
conclusion the following is proven.
Theorem 3.19 ([100]). The Hamiltonian system defined by Hamilton function
(3.57) is not integrable with meromorphic first integrals.
This is the second time we meet the relevance of the logarithmic terms in our
approach; it appeared before in the proof of the non-integrability of the Hénon-
Heiles system using the higher order variational equations in Section 2.4.
Another remark is that this is the only case we know in which the obstruction
to integrability is obtained by an integrable first order variational equation. From
Picard-Vessiot theory (Section 2.2), we knew that the reason for this behavior is
that the identity component of the Galois group is solvable but not commutative.
In particular, it is easy to see that Ziglin’s method does not work in this problem,
essentially because the identity component of the Galois group is unipotent.
For more details about the non-integrability of Hill’s problem see the original
reference [100].
In paper [118] a proof is given of the non-integrability of Hill’s problem by
means of algebraic first integrals.
3.5. Other N-bodiy Problems, Mechanical and Physical Problems.

We briefly survey some works (with references) by several authors about other
Celestial Mechanics problems, (non-Celestial) N- body problems and some physical
problems using our main Theorems of Section 2.3 and Section 3.2. Although some
of them are not related to the main topics in this section, we decided to include
them here.
a) Satellite’s and asteroid’s dynamics. In some Celestial Mechanics problems,
such as satellite’s dynamics, it is important to take into account the shape of the
bodies. If the potential is expanded in harmonics and we truncate it, we may ask
about the non-integrability of these truncated potentials.
In the same way, a system arising from truncation of the harmonics was con-
sidered by Arribas and Elipe in [7]. These authors consider the motion of a particle
around a massive straight segment; for instance, this system can approximatively
represent the dynamics around some asteroids. The potential is of logarithmic type
and homogeneous of degree k = 0. Thus, we cannot apply Theorem 3.1. Then the
188
186
potential is expanded in harmonics and, as above, by truncation it is possible to ap-

ply the Mondéjar result, Theorem 3.4, and the truncated system is non-integrable.
We remark that, without any further analysis, from the non-integrability of the
truncated potential does not imply the non-integrability of the original potential.
For instance, non-integrable systems are obtained by truncation of an expansion of
an integrable Toda lattice [147].
Another sort of problems connected to satellite dynamics comes from consid-
ering the satellite as a finite rigid body rather than as a point particle. It is
well-known that when the three principal inertial moments of the satellite are the
same the system is integrable. Under the assumption that the satellite has axial
symmetry, i.e., that two of the inertia moments are equal, Audin and Boucher stud-
ied the non-integrability of this system using a previous result by Boucher and Weil
about the non-commutativity of the identity component of the Galois group of the
normal variational equations in presence of formal logarithmic branching points in
irreducible factors (see the Appendix A, Section A.3) ([18, 10]); in the proof they
used the computation of the normal variational equation along a suitable integral
curve previously obtained by Maciejewski (unpublished). Maciejewski and Przy-
bylska studied the non-integrability of this system by considering also the action of
the Earth’s magnetic field ([70]).
b) Other N-bodies, Mechanical and Physical problems. Several perturba-
tions of the integrable hydrogen atom are considered by Ferrer and Mondéjar. The
Zeeman Hamiltonian is given by a perturbation of the atom with a magnetic field.
Their non-integrability was shown by Kummer and Sáenz in [62] by using Ziglin’s
theorem. The Stark-Zeeman Hamiltonian is a generalization of the Zeeman Hamil-
tonian obtained through the additional consideration of action of an electric field
parallel to the magnetic field. The non-integrability of the Stark-Zeeman model by
rational first integrals was proven in [38] using Theorem 2.8. The non-integrability
of the hydrogen atom under a combination of magnetic field and circularly polar-
ized microwaves was studied in [39], and a proof of the non-integrability of the
generalized van der Waals Hamiltonian was given in [40].
The non-integrability of several Three-body Problems in atomic physics is stud-
ied by Almeida, Stuchi and López-Castillo ([4, 122]. In the non–integrability proof
they used Theorem 3.1, but we believe that it would be an interesting exercise for
the interested reader to go further and obtain the non-existence of an additional
meromorphic first integral using Theorem 3.3 as we did in Subsection 3.4 for the
Celestial Mechanics case. In fact, in [122] three of the eigenvalues of the Hessian
matrix along a suitable homothetical point are −2, 1, 1; they come from the classi-
cal first integrals in involution: the energy and the two components of the angular
momentum, respectively. The other three eigenvalues depend on the parameters
(masses and atomic number) of the system.
The Gross-Neveu systems are classical Hamiltonian systems with n-degrees of
freedom given by a potential linked to the root system of a simple Lie algebra

V (x1 , . . . , xn ) = exp(< α, x >),
α
where x = (x1 , . . . , xn ) and α = (α1 , . . . , αn ) moves along the root system. These
systems can be considerered as certain generalizations of the Toda lattice. Ma-
ciejewski, Przybylska and Stachowiak proved in [76] the non–integrability of the
187
47
Gross-Neveu systems when the associated Lie algebras are so(2n), so(2n + 1),
sl(n + 1) and sp(2n) (n ≥ 2).
Maciejewski and Przybyska completely solved the integrability problem of a
generalized two–fixed–centres–problem whose interaction potential is V = −ar −2n .
This is a generalization of the classical two fixed centres problem of Celestial
Mechanics for the Newtonian potential. The system is integrable if and only if
n ∈ {−2, −1, 0, 1/2} ([73]).
The integrability of the spring–pendulum system was studied by Churchill,
Delgado and Rod in [26]. It was also studied by the authors in [99], where we
completely solved the problem for physical values of the mass parameter . A gen-
eralization was considered in [78].
The integrability of the dynamical problem of the geodesic on an ellipsoid
was proven by Jacobi. Bardin, Maciejewski and Przybylska completely solved the
integrability of a generalization of this problem, when the particle moves on the
ellipsoid a1 x2i + a2 x22 + a3 x23 = 1 under the action of a quadratic homogeneous
potential V = 1/2(b1 x2i + b2 x22 + b3 x23 ) ([14]).
The Stormer problem is a Hamiltonian system representing the dynamics of a
charged particle in a magnetic dipole. It is a model of the dynamics of charged
particles under the action of Earth’s magnetic field. The non-integrability of the
Stormer problem was proven by Kummer and Sáenz by means of Ziglin’s Theorem
in [63]. The non-integrability of an anisotropic Stormer problem was studied by
Almeida and Stuchi in [4].Furthermore, Sáenz in [117] proved the non-integrability
of another problem of magnetic confinement. The points at infinity are irregular
singularities; in fact, although it was not explicitly stated in [117], by means of
Theorem 2.8, Sáenz proved the obstruction to the existence of rational first integrals.
Other anisotropic families of potentials generalizing the anisotropic Kepler
problem, such as the anisotropic Maneff problem, are studied in [8] by Arribas,
Elipe and Riaguas.
4. Hamiltonian Rigid Body Problem

In 1983 using Theorem 1.1 Ziglin completely solved the integrability problem
of the heavy rigid body problem by means of complex meromorphic first integrals
([149]), and in 1997 the integrability by means of real meromorphic first integrals.
In 2005 Maciejewski and Przybylska revisited the Ziglin results at the light of
Theorem 2.7 and recover Ziglin’s results in a simpler way in the Ziglin original
papers ([74]). We give here a necessarily brief account of this nice work. For space
limitations we only survey the first part of the work, i.e., integrability by means of
complex meromorphic first integrals. It is apparent that in order to recover Ziglin’s
result the authors used several of the methods described in our text. This gives an
idea of the need of attacking a non-trivial problem from several points of view.
4.1. The equations of motion . The Euler-Poisson equations governing the

dynamics of a rigid body with a fixed point within a constant gravitational field, in
the moving frame of coordinates attached to the principal axes of inertia, are given
by
(4.1) Ṁ + Ω × M = µk × l, k̇ = k × Ω,
190
188
being M = (p, q, r) the angular momentum, Ω = (p/A, q/B, r/C) the angular
velocity, (A, B, C) the principal momenta of inertia, k = (α, β, γ) the unitary ver-
tical vector, l = (x0 , y0 , z0 ) the unitary vector with origin at the fixed point and
pointed towards the center of gravity and µ the weight of the body multiplied by
the distance from the fixed point to the center of masses.
The system (4.1) is a dynamical system defined over C6 . It has five free real
parameters (l, A, B, C), l · l = 1 and three classical first integrals:
i) The energy integral:
1
H= M · Ω + µk · l.
2
ii) The geometrical integral:
f1 = k · k = 1.
iii) The vertical angular momentum:

f2 = M · k.
Using these integrals it is possible to reduce the system (4.1) to a two degrees of
freedom Hamiltonian system, see [61], pp. 31-32. From a modern point of view this
reduction is studied in the context of the Poisson actions of Lie groups on symplectic
manifolds and the associated momentum map, see for instance [6], Appendix 5. In
fact, the Euler-Poisson equations (4.1) can be considered as a Hamiltonian system
with three degrees of freedom, whose configuration space is the three dimensional
rotation group and with the additional first integral f2 ; in this case, the geometric
integral f1 is automatically equal to one. Let us denote
Mm = {(M, k) ∈ C6 : f2 (M, k) = m, f1 (M, k) = 1}.

On each level manifold
Mm ,
H|Mm defines a two–degrees of freedom Hamiltonian system. We say that the Euler
equation is (globally) integrable if there exits an additional independent (of H and
f2 ) first integral f3 globally defined in the five–dimensional manifold
M = {(M, k) ∈ C6 : f1 (M, k) = 1};
in this case the Hamiltonian systems defined by H|Mm are integrable in Liouville’s
sense and we can apply our methods.
We remark that from our variational equations point of view it is not necessary
to perform the reduction in this non-linear way, since, at the level of the variational
equations, the reduction is reflected in the obtention of the normal variational
equations N V E.
The known cases of integrability of the equations (4.1) are:
1. (Euler 1758). In the Euler case either there is no gravity (µ = 0 ) or the fixed
point is at the center of masses (l = 0).
2. (Lagrange 1788). The body is axially symmetric, A = B, and the fixed point
belongs to the symmetry axis, x0 = y0 = 0.
3. (Kovalevskaya 1889). For over a century the only cases of integrability known for
the Euler-Poisson equations were those of Euler and Lagrange, but in a celebrated
189
49
paper Sophie Kovalevskaya obtained a new and highly non-trivial case of integra-
bility for A = B = 2C and z0 = 0 ([60]). This paper is the seminal paper that
motivated the actual theory of algebraically completely integrable systems; for the
historical transcendence of this paper see the interesting article of Michèle Audin
[12].
4. (Goryachev-Chaplygin 1910). Under some conditions, a new case of (partial)
integrability was obtained only for m = 0, i.e., if z0 = 0 and A = B = 4C, the
Hamiltonian system defined by H|M0 is integrable in the sense of Liouville ([49]).
4.2. Non–integrability. Using Theorem 1.1, Ziglin proved the following non–
integrability result which solved the integrability problem of the heavy top by means
of meromorphic first integrals.
Theorem 4.1 ([149]). The two–degrees of freedom Hamiltonian system defined
by H|M0 on the zero level symplectic manifold of the vertical angular momentum
f2 , M0 , is integrable by meromorphic first integrals only in the classical integrable
cases (i)–(iv).
As said before, Maciejewski and Przybylska obtained a new simpler proof of
Theorem 4.1 using Theorem 2.7. Their proof follows the following lines (parallel to
the Ziglin’s original proof):
(1) There exists a family of particular pendulum–like solutions Γk of the
Hamiltonian H|M0 .
(2) It is possible to normalize the five parameters of the system in such a way
that y0 = 0 and B = 1.
(3) Using a change of variables (i.e., algebrization procedure, Section A.2 ),
the normal variational equations, NVE, along the family of solutions Γk
are reduced to a second order equation with rational coefficients with four
finite singularities and being the point at infinity an apparent singularity,
provided the parameter
1 1 1
d= ( − )x0 z0
2 C A
is different from zero. In this case using Kovacic’s algorithm and the in-
variance of the identity component by a finite ramified covering (Appendix
A, Theorem A.1), the authors proved that the identity component of the
Galois group of the NVE is not commutative.
(4) When d = 0 and we do not consider the Lagrange integrability case (ii),
z0 = 0 and the above N V E degenerate trough confluence into an hyperge-
ometric equation and applying Kimura’s theorem (Appendix B, Theorem
B.1) only five possible discrete families for the values of the parameter C
are compatible with the commutativity of the identity component of the
Galois group of this hypergeometric equation.
(5) For z0 = 0 there is another family of particular elliptic solution curves Γ1k ,
such that the normal variational equation along them, N V E1 , are reduced
to a family of Lamé type equations (see Appendix C):
d2 ξ e
(4.2) − ((2C(2C + 1)℘(t) + C(1 − 4C))ξ = 0,
dt2 3
192
190
where e = 2k2 − 1, (k ∈ (0, 1) is the moduli). Using Corollary C.2 of

Appendix C and Dwork’s result (Proposition C.3 of Appendix C), it is
possible to prove that the only values of C for which there is a commutative
identity component in the Galois group (of N V E1 ) for the previously
obtained five discrete families are C = 1, C = 1/2 and C = 1/4, which
correspond to Euler, Kovalevskaya and Goryachev-Chaplygin integrable
cases, respectively.
5. Cosmological Models
Some years ago we proved the non-integrability of the Bianchi IX model by
means of rational first integrals. Along these years other non-integrability results
of some cosmological models were obtained by several authors. We survey these
results.
5.1. Bianchi’s Models. The Bianchi cosmological models are a family of rel-
ativistic homogeneous anisotropic models (in the spatial variables) reducing to finite
dimensional Hamiltonian systems. In this section we sketch the non-integrability
proof of two members of this family: Bianchi IX and Bianchi VIII. For the obtention
of the Bianchi family of models, see [65].
The non-integrability proof of the Bianchi IX model was obtained in [99] (see
also [95]); we follow this reference. The Bianchi IX Cosmological model is a dy-
namical system given by the equations in “logarithmic” time ([65]),
d2 log x1
= (x2 − x3 )2 − x21 ,
dt2
d2 log x2
(5.1) = (x3 − x1 )2 − x22 ,
dt2
d2 log x3
= (x1 − x2 )2 − x23 ,
dt2
with the energy constraint (from physical considerations)
ẋ1 ẋ2 ẋ2 ẋ3 ẋ3 ẋ1

E = −( + + ) + x21 + x22 + x23 − 2(x1 x2 + x2 x3 + x3 x1 ) = 0.
x1 x2 x2 x3 x3 x1
Thus, we get a dynamical system of dimension five on the zero level energy
manifold M0 .
In fact this is a Hamiltonian system with position variables x1 , x2 , x3 and con-
jugate moments given by
1 d
y1 = − log(x2 x3 ),
x1 dt
1 d
y2 = − log(x1 x3 ),
x2 dt
1 d
y3 = − log(x1 x2 ).
x3 dt
Hence, the energy becomes the two degrees of freedom Hamiltonian
1
HIX = (x21 y12 + x22 y22 + x23 y32 − 2x1 x2 y1 y2 − 2x2 x3 y2 y3 − 2x1 x3 y1 y3 )
4
191
51
+x21 + x22 + x23 − 2x1 x2 − 2x2 x3 − 2x1 x3 = 0.

Our proof of the non-integrability relies on the study of the variational equations
along particular solutions of the Taub family.
As Taub noticed, the subspace x2 = x3 (ẋ2 = ẋ3 ) is invariant by the flow
of the system, and the reduction to this invariant four dimensional space (three
dimensional if we consider the restriction of the system on the restricted manifold
M0 ) is completely integrable (i.e. it is an integrable subsystem T ) and the solutions
can be calculated explicitly ([130], p. 481). From the Taub family of solutions we
select the particular ones
2k k cosh(2kt)
x1 = , x 2 = x3 = ,
cosh(2kt) 2 cosh2 (kt)
being k a parameter. This particular integral curve (for a fixed value of k) is our
integral curve and is contained in M0 .
Along the above particular integral curve (for a fixed value of k) we compute
the variational equation and the normal variational equation, NVE. The NVE can
be reduced to a second order differential equation with rational coefficients
1 1 5 1 3 1
(5.2) η̈ + ( + + )η = 0.
4 x − 1 4 (x − 1)2 16 x2
This equation has x = 0, 1 as regular singularities, being x = ∞ an irregular
singularity; in fact, it is a confluent Heun’s equation, i.e., an equation obtained from
a confluence of two singular points into an equation with four regular singularities
on the Riemann sphere. For the physical meaning of these singular points, see [99]
(or [95]).
Then by means of our Theorem 2.8 and using Kovacic’s algorithm (Appendix
A, Section A.1), we proved in [99] the non-integrability of this system by means
of rational first integrals. We point out that we will use Theorem 2.8, instead of
Theorem 2.7, because the points at ∞ of the particular integral curve in phase
space correspond to an irregular singularity of the NVE (see the original reference
[99], or the book [95]), for details).
In an analogous way, Maciejewski, Strelcyn and Szydlowski proved in 2001 the
non-integrability of the Bianchi VIII cosmological model ([80]). The Hamiltonian
is now
HV III = −x22 y12 − x21 (1 + y22 ) − 2x1 y1 (x2 y2 − x3 y3 )
1
+x2 x3 (2y2 y3 − 1) − x23 (1 + 4y32 ).
4
Inasmuch as for Bianchi IX, the Hamiltonian HV III restricted on the Taub
manifold X1 = 0, x1 = 0 is an integrable subsystem, which on the five dimensional
zero energy level manifold, M0 = HV−1III (0), gives rise to a three dimensional inte-
grable (Taub) subsystem T . Then the authors in [80] integrate the subsystem T
and compute the NVE along this three-parametric family of particular solutions.As
in HIX it is possible to reduce these variational equations to second order varia-
tional equations with rational coefficients and with our Theorem 2.7 and Kovacic’s
algorithm they proved the non-integrability of the Hamiltonian system of HV III
194
192
by means of meromorphic first integrals. Independently of the analogies, there two

main differences here with respect to our previous result on HIX :
(1) In [80] is computed the NVE along the complete three parameter family
defined by the Taub subsystem on M0 , not only through a one parameter
family as was the case for [99]. This implies that the authors obtain
the non-integrability of the Hamiltonian system defined by HV III in a
neighborhood of the Taub family of solutions restricted to M0 .
(2) The NVE in [80] have regular singular points at ∞, for this reason it is
possible to apply Theorem 2.7 and to find obstruction to the existence of
meromorphic first integrals instead of rational ones.
5.2. Friedman-Robertson-Walker’s Models . The Friedman-Robertson-

Walker (FRW) cosmological models are the classical relativistic homogeneous isotropic
models that also can be reduced to finite dimensional Hamiltonian systems (see
[67]).
One important FRW family of models with a conformally coupled self-interacting
scalar field is defined by the two degrees of freedom Hamiltonian
1 1 Λ λ
HF RW = (−y12 + y22 ) + k(x22 − x21 ) + m2 x21 x22 + x41 + x41 ,
2 2 2 2
where k ∈ {−1, 0, 1} is the spatial curvature and m, Λ, λ are real parameters which
represent the mass of the scalar field, the cosmological constant and the self-coupling
constant, respectively.
The first application of the methods in this text to the non-integrability of
a FRW model was obtained by Maciejewski and Szydlowski in 2000 as follows
([67]). The flat subfamily with k = 0 in the family HF RW gives rise through the
complex canonical change (x1 , y1 ) → (−ix1 , iy1 ) to a classical Hamiltonian with an
homogeneous potential of degree four. Thus, it is possible to apply Theorem 3.1
and to obtain necessary conditions for integrability: if the parameters m, Λ, and λ
do not satisfy some concrete discrete families of algebraic relations, the system is
not integrable with meromorphic first integrals.
Coelho, Skea and Stuchi also studied in [28] the non-integrability of the family
HF RW but now with k = 0. As x1 = y1 = 0 is an invariant plane, it gives rise to
a family of particular solutions parametrized by the energy h, like in Section 3.3.
The NVE along this family is a Lamé type equation (Appendix C) that the authors
write in algebraic form and by Theorem 2.7 using Kovacic’s algorithm (Appendix
A) the authors proved that a necessary condition for meromorphic integrability is
that
2m2
(5.3) {λ, Λ} ⊂ {− : p ∈ N}
(p + 1)(p + 2)
Very recently Boucher and Weil continued the work of [28] and obtained
stronger restrictions on the parameters compatible with the integrability of HF RW
([20]). These authors systematically used the obstruction to the integrability given
by the existence of logarithmic terms either in the first variational equation or in
higher order variational equations. In particular, they recovered the conditions (5.3)
for integrability applying their own criterium to the first order NVE (see Appendix
193
53
A, Section A.3) and, as the first order variational equation is given by a two uncou-
pled Lamé-Hermite equations (A.8), by means of Proposition A.6 of Appendix A
these authors obtain other restrictions for meromorphic integrability of the Hamil-
tonian system defined by HF RW . Moreover, Boucher and Weil conjectured that
this system is integrable if, and only if, either λ = Λ = −m2 or λ = Λ = −m2 /3.
In another recent paper, Maciejewski, Przybylska, Stachowiak and Szydlowski
studied the integrability of the FRW cosmological model defined by the Hamiltonian
1 y2
(5.4) H= − y12 + 22 − kx21 + Lx41 + m2 x41 x22 .
2 x1
This Hamiltonian represents a FRW cosmological model with a complex scalar
field. As with the Hamiltonian HF RW , k and m are the curvature and the mass
of the field, respectively, being L essentially (i.e., modulo a constant factor) the
cosmological constant. For L = 0 the authors obtain an algebraic form of a NVE of
Whittaker type, i.e., a confluent hypergeometric equation with a regular singular
point at the origin and an irregular one at the infinity and, using a result of the sec-
ond author about the Galois group of this kind of equations (see [86]), by Theorem
2.8 a necessary condition is obtained for integrability with rational first integrals:
namely, that the curvature k and energy h must be zero. For L = 0 and h = 0 then
either k = 0 or 9 − 4m2 /L = (2p + 1)2 , for some p ∈ Z, using Kimura’s theorem
(Theorem B.1 of Appendix B) and Theorem 2.7, because the NVE is reduced to
an hypergeometric equation, equation (B.2) of Appendix B. For L = 0 and h = 0
(i.e., we are on the physical manifold M0 ) then either k = 0 or 9 − 4m2 /L = 4p)2 ,
for some p ∈ Z, using again Proposition A.6 of Appendix A.
Although it is not directly connected with the main body of this text, we
mention here the recent work in [127], where the differential Galois theory is applied
directly (i.e., without use of the variational equations) to the integrability in closed
form of the differential equations obtained as linear density perturbations of FRW
model. The paper is a nice and relatively simple example of another application
of the differential Galois theory to integrability problems. The main technical tool
used by the authors was the Kovacic algorithm together with the algebrization
mechanism (Section A.2 of Appendix A).
6. An Application to Painlevé’s Transcendents

The Painlevé transcendents are the solutions of the six Painlevé’s families of
equations (P1 , P2 , P3 , P4 , P5 , P6 ). Since the work of Malmquist we know that the
Painlevé transcendents can be expressed as Hamiltonian systems of 1 + 1/2 degrees
of freedom.
We conjecture that all Painlevé equations are non-integrable by means of ratio-
nal first integrals. Using the fundamental theorems of subsections 2.3 and 2.4 it is
possible to prove this result for some particular cases (for some particular equations
and some particular values of the parameters); in the state of the art it works for
some special values of the parameters in the cases of P1 and P6 , as we shall explain
below. Moreover, using an extension of our main theorems, replacing differential
Galois groups by differential Galois groupoids, thus linear differential Galois the-
ory by non-linear differential Galois theory, it is possible to prove that P1 and P6
(for all the values of the parameters) are non-integrable by means of rational first
integrals.
196
194
In the cases of P1 , P2 , P3 , P4 , P5 , as the variational equation along each

particular solution will have an irregular singular point at infinity, then, using
theorem 2.8 (or its extension to higher order variational equations), we can only
hope to obtain obstruction to the integrability by means of rational first integrals.
In the case of P6 it is possible to obtain an obstruction to the integrability by means
of meromorphic first integrals for some special values of the parameters.
Here we illustrate our approach with Painlevé II and Painlevé VI, but, of course,
we believe that similar studies can be done for other Painlevé’s families with rational
particular solutions.
6.1. Painlevé II. We shall obtain a non-integrability result for a discrete

subfamily of Painlevé II equations (cf. also [2, 129]). We follow [101], to which
we address the reader for more details and remarks.
The second Painlevé transcendent is given by the solutions of the Painlevé II
equation
(6.1) ẍ = 2x3 + tx + α,
α being a complex parameter.
For Painlevé II the Hamiltonian is
1 1 1
H0 (y, x, t) = y 2 − (x2 + t)y − (α + )x,
2 2 2
and the differential equation (6.1) is equivalent to the Hamiltonian system
∂H0 1 ∂H0 1
ẋ = = y − x2 − t, ẏ = − = 2xy + α +
∂y 2 ∂x 2
([85, 107]).
Now, by a standard procedure in Hamiltonian dynamics, from the above non-
autonomous Hamiltonian system we can obtain a two degrees of freedom autonomous
Hamiltonian system such that the non-autonomous system is included as a subsys-
tem. For the Hamiltonian H0 , it is given by
H(y, x, z, e) = H0 (x, y, z) + e.
Thus, the associated Hamiltonian system is
ẋ = y − x2 − 12 z,
ẏ = 2xy + α + 12 ,
(6.2)
ż = 1,
ė = 12 y.
It seems clear that the dynamical system (6.2) is equivalent to the Painlevé
II equation (6.1), in the sense that from the solutions of one we can immediately
obtain the solutions of the other. In particular, for any reasonable meaning of the
word “integrable”, the integrability of one of them implies the integrability of the
other. We remark that the function e(t) = 12 y(t)dt is closely related to the τ
function of the Painlevé equation (6.1) ([108]).
The variational equation along Γ : x = x(t), y = y(t), z = z(t), e = e(t) is
195
55
⎛ ⎞ ⎛ ⎞⎛ ⎞
ξ1 −2x(t) 1 − 12 0 ξ1
⎜ ⎟ ⎜
d ⎜ ξ2 ⎟ ⎜ 2y(t) 2x(t) 0 0 ⎟ ⎜ ξ2 ⎟
(6.3) = ⎟⎜ ⎟.
dt ⎝ ξ3 ⎠ ⎝ 0 0 0 0 ⎠ ⎝ ξ3 ⎠
1
ξ4 0 2 0 0 ξ4
The normal variational equation is given by

d ξ1 −2x(t) 1 ξ1
(6.4) = .
dt ξ2 2y(t) 2x(t) ξ2
Given a differential system

d ξ1 a(t) b(t) ξ1
(6.5) = ,
dt ξ2 c(t)) d(t)) ξ2
with coefficients in a differential field K, by an elimination process it is equivalent
to the second order equation
ḃ(t) a(t)ḃ(t)
(6.6) ξ¨ − (a(t) + d(t) + )ξ̇ − (ȧ(t) + b(t)c(t) − a(t)d(t) − )ξ = 0,
b(t) b(t)
where ξ := ξ1 . We remark that the equations (6.5) and (6.6) are equivalent in the
sense that they represent the same D-module (see [139]). In particular, the Galois
groups of both equations are the same.
Hence the normal variational equation (6.4) is equivalent to the second order
equation
(6.7) ξ¨ − (2y(t) − 2ẋ(t) + 4x2 (t))ξ = 0.

Now by using the Hamilton equations (6.2) and taking z(t) = t, we obtain
(6.8) ξ¨ − (6x2 (t) + t)ξ = 0.
Now we fix α = 1. Then it is well-known that the equation (6.1) has the
particular solution (see, for instance, [41])
1
(6.9) x=−
t
and the associated Hamiltonian system (6.2) has the particular rational solution
1 2 t 1 t2
(6.10) Γ : x(t) = − , y(t) = 2 + , z(t) = t, e(t) = − + .
t t 2 t 8
For this particular solution, (6.8) is given by
6
(6.11) ξ¨ − ( 2 + t)ξ = 0.
t
By means of the change of variable ξ(t) = t1/2 η(x), x = i 23 t3/2 , it is transformed
into Bessel’s equation
198
196
d2 η dη
(6.12) x2 +x + (x2 − n2 )η = 0,
dx2 dx
with n = 5/3.
Now it is well-known that when n ∈/ Z + 1/2 the identity component of Galois
group of Bessel’s equation is non-commutative, indeed, for these values the Galois
group is SL(2, C) (see, [95], Subsection 2.8.2, for a simple proof using Stokes ma-
trices). As the point at z = t = ∞ is an irregular singular point of the variational
equation, by Theorem 2.8, we have proven the following proposition:
Proposition 6.1. For α = 1, the Hamiltonian system (6.2) associated to the
Painlevé II equation is not integrable by means of rational first integrals.
Furthermore, it a classical fact that not only for α = 1, but for any integer α
the Painlevé II equation has rational particular solutions (such a solution is (6.9)
for α = 1) and there are rational changes of variables in the phase variables called
Bäcklund (or canonical) transformations between the members of this discrete fam-
ily of Hamiltonian systems ([41, 108]). Hence if one of them is non-integrable by
rational first integrals, any member of this family satisfies the same property. We
have proven the following:
Corollary 6.2. For α ∈ Z, the Hamiltonian system (6.2) associated to the
Painlevé II equation is not integrable by means of rational first integrals.
6.2. Painlevé VI. The Painlevé VI transcendent is given by the solutions of
the Painlevé VI equation
2 1
ẍ = 12 x1 + x−1
1 1
+ x−t ẏ − t + t−1
1 1
+ x−t ẏ
(6.13) x(x−1)(x−t) t (t−1) t(t−1)
+ t2 (t−1)2 α + β x2 + γ (y−1)2 + δ (y−t)2
α, β, γ, δ being complex parameters.
There are other ways to write the parameters, more natural in the interpretation
2
θ2
of Painlevé VI as related to an isomonodromic deformation: α = (θ4 −1)
2 , β = − 21 ,
θ22 1−θ 2
γ= 2
= 2 3.
,δ
The case α = β = γ = 0, δ = 12 (θ1 = θ2 = θ3 = 0, θ4 = 1) was studied by E.
Picard before Painlevé discovery of Painlevé equations, it is called Picard-Painlevé
case: P P6 .
In [46] the authors proved that Painlevé VI with parameters α = β = γ =
δ = 0 (and more generally with parameters related by Backlünd transformations)
is non integrable by means of meromorphic first integrals. We will only sketch
their proof and we address the reader to [46] for more details and remarks. In
this case it is necessary to use the second variational equation and the authors
obtain an obstruction using a special function, the dilogarithm. More generally
we can think that in the case of an “elementary” first variational equation with
fuchsian logarithmic singularities only at 0, 1, ∞ ∈ P1 and an abelian monodromy,
the polylogarithms could be an efficient tool to obtain an obstruction to integrability
by means of higher variational equations.
We suppose α = β = γ = δ = 0. Then (6.13) is equivalent to the Hamiltonian
system
197
57
2x(x − 1)(x − t)
ẋ = y
t(t − 1)
y2
ẏ = − (x − 1)(x − t) + x(x − 1) + x(x − t)
(6.14) t(t − 1)
ṫ = 1
1 2t − 1
ė = x(x − 1)y 2 + 2 x(x − 1)(x − t)y 2 .
t(t − 1) t (t − 1)2
This system possesses a simple family of solutions x = c, y = 0, e = E, where
c, E are constants. The first normal variational equation is
2c(1 − c)(c − t)
ξ˙1 = ξ2
(6.15) t(1 − t)
ξ˙2 = 0
Choosing ξ2 = 1, and setting C = c(1−c), we get ξ1 = 2C c ln(t)+(1−c) ln(1−

t) . The corresponding differential Galois group is commutative; thus, there is no
obstruction to integrability at the first order. We remark that this Galois group is
connected, therefore the Galois groups of the higher variational equations are also
connected.
The second order component of the second variational equation (in its non
linear form) is
(2) c 1 − c ˙(2) c(3c − 2) (1 − c)(3c − 1)

ξ˙1 = 2C − ξ +2 − + ξ̇1 ξ̇2
t 1−t 2 t 1−t
(6.16)
(2) c(3c − 2) (1 − c)(3c − 1) 2
ξ˙2 = − ξ̇2
t 1−t
We set A0 = c(3c−2), A1 = (1−c)(1−3c), then the linear version of the second
variational equation is given by the first variational equation and the following
equation:
⎛ (2) ⎞ ⎛ c
A0 A1
⎞ ⎛ (2) ⎞
ξ1 0 2C t − 1−c
1−t −2 t + 1−t 0 ξ1
d ⎜ (2) ⎟ ⎜ A0 A1 ⎟ ⎜ξ (2) ⎟
(6.17) ⎜ξ1 ⎟ = ⎜0 0 0 t + 1−t ⎟ ⎜ 1 ⎟
dt ⎝ u ⎠ ⎝0 0 0 2C ct − 1−c
1−t
⎠⎝ u ⎠
v 0 0 0 0 v
where u := ξ1 ξ2 , v := ξ22 .
t
We introduce the dilogarithm Li2 (t) = − 0 ln(1−s)s ds. Using the monodromies
of Li2 (t) and Li2 (1 − t) it is possible to compute two generators of the monodromy
of (6.17) (in a convenient basis), we obtain
⎛ ⎞
1 4πiCc −4πiA0 4π 2 A0 Cc
⎜0 1 0 2πiA0 ⎟
(6.18) M0 := ⎜
⎝0
⎟
0 1 4πiCc ⎠
0 0 0 1
corresponding to a loop around 0 and
200
198
⎛ ⎞
1 −4πiC(1 − c) −4πiA1 −4π 2 A1 C(1 − c)
⎜0 1 0 2πiA1 ⎟
(6.19) M1 := ⎜
⎝0
⎟
0 1 −4πiCc(1 − c) ⎠
0 0 0 1
corresponding to a loop around 1.
The matrices M0 and M1 do not commute for a generic value of c. By theorem
2.10, this gives us an obstruction to integrability by means of meromorphic first
integrals, because the Galois group of the second variational equation is connected.
6.3. More about the non integrability of Painlevé I and Painlevé VI.
There is a non-linear version of differential Galois theory due independently to B.
Malgrange [83] and H. Umemura [138]. J. Drach tried to built such a theory in
his thesis but there are some gaps [31]. We will use Malgrange version, because
its geometric approach is more adapted to our purposes. Since the non-linear
differential Galois theory is quite technical we will only sketch some ideas without
precise definitions, addressing the reader to the original papers for more details.
We recall that a groupoid is a small category whose all the morphisms are
isomorphisms. Malgrange introduced Lie D-groupoids (we will say Lie groupoids
for simplicity); roughly speaking they are the subgroupoids of the groupoid of germs
of analytic diffeomorphisms of an analytic complex manifold M defined by systems
of analytic PDE. A Lie groupoid has a Lie algebra. There is also an algebraic
version when M is an algebraic manifold.
By definition, the Galois groupoid of an analytic singular foliation is the small-
est Lie groupoid among the Lie groupoid whose Lie algebra “contains the tangent
pseudogroup of the foliation”. Given a system, its Galois groupoid is the Galois
groupoid of the corresponding foliation; it is defined on the phase space in the
autonomous case and on the extended phase space in the general case.
There is a non linear version of theorem 2.10 when one replaces the usual
differential Galois theory by the non-linear Galois theory (it is due to the second
author, cf. [102], part 5, p. 27): if a Hamiltonian system is integrable, then the
Lie algebra of its non linear Galois groupoid is abelian.
For Painlevé’s equations the Galois groupoid is defined on the extended phase
space C3 and it always preserves a non trivial closed rational 2-form ω = iX dt ∧
dx ∧ dy, where X is the vector field on C3 associated to the equation (which is
divergence free).
Painlevé I is the differential equation:
(6.20) ẍ = 6x2 + t.
∂ ∂ ∂
The associated vector field on the phase space C3 is X = ∂t +y ∂x +(6x2 +t) ∂y .
In this case the preserved form is ω = iX dt ∧ dx ∧ dy. We have the following result
due to G. Casale [24].
Theorem 6.3. The Galois groupoid of Painlevé I is the algebraic Lie groupoid
on the phase space preserving the form ω = iX dt ∧ dx ∧ dy. Its solutions are the
germs of transformations f of C3 such that f ∗ ω = ω.
P. Painlevé and J. Drach “proved” a similar result but there were gaps and
errors in their proofs.
201
59
Therefore the Lie algebra of the Galois groupoid of Painlevé I is non abelian
and, applying the non linear version of our theorem, we obtain the following result.
Theorem 6.4. Painlevé I is not integrable by means of rational functions.
The following result is due to S. Cantat and F. Loray [66] (the proof uses deep
results of G. Casale on non-linear Galois theory and of the japanese school on the
dynamics of Painlevé VI [51]).
Theorem 6.5. The Galois groupoid of Painlevé VI is the algebraic Lie groupoid
on the phase space preserving the form ω = iX dt ∧ dx ∧ dy, except in each one of
the cases:
• θj ∈ 12 + Z, j = 1, 2, 3, 4;
• θj ∈ Z, j = 1, 2, 3, 4 and θ1 + θ2 + θ3 + θ4 is odd.
All these cases are equivalent modulo Okamoto symmetries to the case of P P 6:
θ1 = θ2 = θ3 = 0, θ4 = 1.
The following result is due to G. Casale [24].

Theorem 6.6. The Galois groupoid of Painlevé-Picard VI is transversally
affine.
Using Okamoto symmetries we can get similar results in each one of the excep-
tional cases of the preceding theorem. Then, in all the cases, exceptional or not,
the Lie algebra of the Galois groupoid of Painlevé VI is non abelian, and, applying
the non linear version of our results, we obtain the following result.
Theorem 6.7. For all values of the parameters Painlevé VI is not integrable
by mean of rational functions.
There is some hope to extend the approach of Cantat and Loray [66], using
cubic surfaces, to P2 , P3 , P4 and P5 , replacing the usual dynamics (linear and
non-linear monodromy) by some “Stokes dynamics” (linear and non-linear Stokes
phenomena).
Using the non linear version of our theorem we can avoid the choice of a par-
ticular solution, which is very good, but unfortunately there is a price to pay: the
proofs and computations are more difficult.
Appendix A. Algorithmic Considerations

A.1. Kovacic Algorithm . For the sake of completeness we include here the
standard Kovacic algorithm, because, as it was shown in the previous pages, it is
still very useful in the applications to non-integrability.
The Kovacic algorithm gives us a procedure in order to compute the Picard-
Vessiot extension (i.e., a fundamental system of solutions) of a second order differ-
ential equation, provided the differential equation is integrable. Reciprocally, if the
differential equation is non-integrable, the algorithm does not work (see[59]). In
this (necessarily brief) description of the algorithm we essentially follow the version
of the algorithm given in [33, 34].
Given a second order linear differential equation with coefficients in C(x), it is
a classical fact that it can be transformed into the so-called reduced invariant form
202
200
(A.1) ξ + gξ = 0,
with g = g(x) ∈ C(x).
We remark that in this change we introduce the exponentiation of a quadrature
and the integrability of the original equation is equivalent to the integrability of
the above equation although, in general, the Galois groups are not the same.
The algorithm is based on the following two general facts:
(A) The classification of the algebraic subgroups of SL(2, C) given in Proposition
2.5 of Section 2.1 (the Galois group of the equation (A.1) is contained in SL(2, C)).
(B) The well-known transformation to a Riccati equation, by the change v = −ξ /ξ,
(A.2) v = g + v2 .
Then (see Section 2.2) the differential equation (A.1) is integrable, if and only if,
the equation (A.2) has an algebraic solution. The key point now is that the degree
n of the associated minimal polynomial Q(v) (with coefficients in C(x)) belongs to
the set
Lmax = {1, 2, 4, 6, 12}.

The determination of the set L of possible values for n, is the First Step of
the algorithm. We remark that for n = 4, n = 6 and n = 12, the Galois group of
(A.1) is finite (hence these values are related to the crystalographic groups). The
two other steps of the algorithm (Second Step and Third Step) are devoted
to computation of the polynomial Q(v) (if it exists). If the algorithm does not
work (i.e., if the equation (A.2) has no algebraic solution) then equation (A.1) is
non-integrable and its Galois group is SL(2, C).
Now we will describe the algorithm.
Let
s(x)
g = g(x) = − ,
t(x)
with s(x), t(x) relatively prime polynomials, and t(x) monic. We define the follow-
ing function h on the set Lmax = {1, 2, 4, 6, 12}, h(1) = 1, h(2) = 4, h(4) = h(6) =
h(12) = 12.
First Step
If t(x) = 1 we put m = 0, else we factorize t(x) in monic relatively prime
polynomials. Then
1.1. Let Γ be the set of roots of t(x) (i.e., the singular points at the finite complex
plane) and let Γ = Γ ∪ ∞ be the set of singular points. Then the order at a
singular point c ∈ Γ is, as usual, o(c) = i if c is a root of multiplicity i of t(x). The
order at infinity is defined by o(∞) = max(0, 4 + deg(s) − deg(t)). We call m+ the
maximum value of the order that appears at the singular points in Γ, and Γi is the
set of singular points of order i ≤ m+ .
1.2. If m+ ≥ 2 then we write γ2 = card(Γ2 ), else γ2 = 0. Then we compute
203
61
*
γ = γ2 + card( Γk ).
k odd
3≤k≤m+
1.3. For the singular points of order one or two, c ∈ Γ2 ∪ Γ1 , we compute the
principal parts of g:
g = αc (x − c)−2 + βc (x − c) + O((x − c)2 ),

if c ∈ Γ , and
g = α∞ x−2 + β∞ x−3 + O(x−4 ),

for the point at infinity.
1.4. We define the subset L (of possible values for the degree of the minimal
polynomial Q(v)) as {1} ⊂ L if γ = γ2 , {2} ⊂ L if γ ≥ 2 and {4, 6, 12} ⊂ L if
m+ ≤ 2.
1.5. We have the three following mutually exclusive cases:
1.5.1. If m+ > 2, then L = L .
1.5.2. If m+ ≤ 2 and the two following conditions are satisfied:
√
1.5.2.1. For any c ∈ Γ, 1 + 4αc ∈ Q, and c∈Γ βc = 0,
√
1.5.2.2. For any c ∈ Γ such that 1 + 4αc ∈ Z, logarithmic terms do not appair
in the local solutions in a neigbourhood of c,
then L = L .
1.5.3. If cases 1.5.1 and 1.5.2 do not hold then L = L − {4, 6, 12}.
1.6. If L = ∅, then equation (A.1) is non-integrable with Galois group SL(2, C),otherwise
one writes n for the minimum value in L.
We remark that condition 1.5.2.2 is not stated in the original Kovacic’s paper.
As the reader can check, it follows trivially from the fact that the existence of a
logarithm in a local solution is an obstruction to a finite monodromy and Galois
group. We decided to include this condition here because it has successfully been
applied to some important Hamiltonian systems ([53, 52]).
For the Second Step and the Third Step of the algorithm we consider the
value of n fixed.
Second Step
2.1. If ∞ has order 0 we write the set
h(n) h(n) h(n) h(n)
E∞ = {0, ,2 ,3 , ..., n }.
n n n n
2.2. If c has order 1, then Ec = {h(n)}.

2.3. If n = 1, for each c of order 2 we define
1 √ 1 √
Ec = { (1 + 1 + 4αc , (1 − 1 + 4αc }.
2 2
204
202
2.4. If n ≥ 2, for each c of order 2, we define
h(n) √ h(n) √
Ec = Z ∩ { (1 − 1 + 4αc ) + k 1 + 4αc : k = 0, 1, ..., n}.
2 n
2.5. If n = 1, for each singular point of even order 2ν, with ν > 1, we compute the
numbers αc and βc defined (up to a sign) by the following conditions:
2.5.1. If c ∈ Γ ,
ν−1

g = {αc (x − c)−ν + µi,c (x − c)−i }2 + βc (x − c)−ν−1 + O((x − c)−ν ),
i=2
and we write
ν−1
√
g c := αc (x − c)−ν + µi,c (x − c)−i .
i=2
2.5.2. If c = ∞,
ν−3

g = {α∞ xν−2 + µi,∞ xi }2 − β∞ xν−3 + O(xν−4 ),
i=0
and we write
ν−3
√
g ∞ := α∞ xν−2 + µi,∞ xi .
i=0
Then for each c as above, we compute
1 βc
Ec = { (ν + ) : = ±1},
2 αc
and the sign function on Ec is defined by
1 βc
sign( (ν + )) = ,
2 αc
being +1 if βc = 0.
2.6. If n = 2, for each c of order ν, with ν ≥ 3, we write Ec = {ν}.
Third Step
For n fixed, we try to obtain elements e = (ec )c∈Γ in the cartesian product
3.1.
c∈Γ Ec , such that:
n

(i) d(e) := n − h(n) c∈Γ ec is a non-negative integer,
(ii) If n = 2 then there is at least one odd number in e.
If no element e is obtained, we select the next value in L and go to the Second
Step, else n is the maximum value in L and the Galois group is SL(2, C) (i.e., the
equation (A.1) is non-integrable).
205
63
3.2. For each family e as above, we try to obtain a rational function Q and a
polynomial P , such that
(i)
n ec √
Q= + δn1 sign(ec ) g c ,
h(n)
x−c
c∈Γ c∈∪ν>1 Γ2ν
where δn1 is the Kronecker delta.

(ii) P is a polynomial of degree d(e) and its coefficients are found as a solution of
the (in general, overdetermined) system of equations
P−1 = 0,
Pi−1 = −(Pi ) − QPi − (n − i)(i + 1)gPi+1 , n ≥ i ≥ 0,
Pn = −P.
If a pair (P, Q) as above is found, then equation (A.1) is integrable and the
Riccati equation (A.2) has an algebraic solution v given by any root v of the equation
n
Pi
v i = 0.
i=0
(n − 1)!
If no pair as above is found we take the next value in L and we go to the
Second Step. If n is the greatest value in L then equation (A.1) is non-integrable
and the Galois group is SL(2, C).
Under some assumptions, in the literature there are other algorithms to com-
pute the Galois group of a linear differential equation. We notice that a remarkable
simplification of the above algorithm was obtained in [137] for irreducible differ-
ential equations. An algorithm for third order differential equations is given in
[125, 126]. For completely reducible equations, i.e. when the Galois group is
reductive, an algorithm is presented in [30].
A.2. Algebrization Procedure . In concrete differential equations it is use-

ful, if possible, to replace the original differential equation over a compact Riemann
surface, by a new differential equation over the Riemann sphere P1 (i.e., with ra-
tional coefficients) by a change of the independent variable. This equation on P1
is called the algebraic form of the equation. In a more general way we will consider
the effect of a finite ramified covering on the Galois group of the original differential
equation.
Theorem A.1. ([97], see also [95]) Let Γ be a Riemann surface,
d
(A.3) ξ = A(x)ξ, A ∈ M at(m, C(x))
dx
a linear differential equation on P1 and x : Γ → P1 , x = x(t) a finite ramified
covering of P1 (t a local parameter in Γ). Let
d
(A.4) ξ = x∗ (A)(t)ξ, x∗ (A) ∈ M at(m, M(Γ))
dt
206
204
be the pull-back of equation (A.3) by x (i.e., the equation obtained by the change
of variables x = x(t)). Then the identity components of the Galois groups of the
equations (A.3) and (A.4) are the same.
We say that a linear differential equation
d
(A.5) ξ = A (t)ξ, A ∈ M at(m, M(Γ))
dt
is algebrizable if it is the pull-back of a linear differential equation (A.3). In order
to apply Kovacic’s algorithm it is important to know whether a given second order
linear differential equation is algebrizable. An algorithm to algebrize equations of
the type
ÿ = r(t)y
is proposed recently in [3].
We say that a change of variable x = x(t) is Hamiltonian if and only if
(x(t), ẋ(t)) is a solution curve of the autonomous 1-degree of freedom Hamilton-
ian system
ẋ2
H = H(x, ẋ) = + V (x).
2
Proposition A.2 (Algebrization algorithm). ([3]) The differential equation
(A.6) ÿ = r(t)y
is algebrizable through a Hamiltonian change of variable x = x(t) if, and only if,
there exist f, α such that
α f
, ∈ C(x), where f (x(t)) = r(t), α(x) = 2(h − V (x)) = ẋ2 .
α α
Furthermore, the algebraic form of the equation ÿ = r(t)y is
1 α f
(A.7) y + y − y = 0,
2α α
where = d/dx.
From the above we know that when r(t) belong to the field of meromorphic
functions over Γ, M(Γ), then the identity component of the Galois group is pre-
served by the above change of variables x = x(t).
A.3. The importance of logarithmic terms . We observe that in the Ko-
vacic algorithm we include a logarithmic condition. This is not an isolated behavior:
the appearance of logarithmic terms in the solution of the variational equations is
very often an obstruction to the integrability. In our opinion this fact gives some
insight about the success of the so-called Kovalewskaya-Painlevé heuristic analysis,
although a complete clarification remains open.
In Section 3.4 we already found that in the Hill Problem the existence of a
logarithmic term was essential for proving its non-integrability. There are other
possible situations where the logarithmic terms are essential.
a) Higher order variational equations. Assume that the first order variational
equation VE 1 of a Hamiltonian system around a particular solution has a commu-
tative identity component of the Galois group G01 . At first order we cannot obtain
obstruction to the integrability. For the moment the only family of systems where
Theorem 2.10 was applied with success is the following; we follow [95], section 8.
207
65
Consider a two degrees of freedom Hamiltonian system with a first order vari-
ational equation, VE 1 , such that is given by the direct sum of two Lamé type
equations (see Appendix C):

ξ¨1 = n1 (n1 + 1) ℘(t) + B1 ξ1 ,
(A.8)
ξ¨2 = n2 (n2 + 1) ℘(t) + B2 ξ2 ,
where the field of coefficients of (A.8), as well as the field of coefficients of the VE k
for k > 1, is a field of elliptic functions K = C(℘(t), ℘(t)),
˙ isomorphic to the field of
meromorphic functions over Γ. For simplicity we consider the case of a first order
variational equation given by only two Lamé type equations but in an obvious way
it can be generalized to an n–degrees–of–freedom Hamiltonian system with a first
order variational equation given by a direct sum of n Lamé type equations. One of
the equations (A.8) is the first order normal variational equation, say the second
one.
Let K ⊂ L1 ⊂ L2 ⊂ · · · Lk the Picard-Vessiot extension when we solve VE k .
From Section 2.3 we know that once the solutions of VE 1 , K ⊂ L1 , are obtained
the solutions of the second order, third order, etc., L1 ⊂ L2 ⊂ L3 · · · are obtained
by the method of variation of constants. So, to get the extension Lk /L1 we only use
quadratures, this extension is a purely transcendental one and by Picard-Vessiot
general theory (Section 2.1) the Galois group Gal(Lk /L1 ) is connected. If the Galois
group G1 is also connected then the extension K ⊂ L1 is also transcendental, the
total Picard-Vessiot extension K ⊂ Lk of VE k is transcendental and Gk = (Gk )0 .
We have proven the following:
Lemma A.3 ([95]). Under the above assumptions, Gk = (Gk )0 if and only if
G1 = (G1 )0 .
The variational equation VE 1 is Fuchsian. Therefore all higher order varia-
tional equations VE k (more precisely, their linear counterparts) are also Fuchsian
(for k ≥ 1, see [102] for the details) and Gk is the Zariski closure of their mon-
odromy groups.
We have the following lemma.
Lemma A.4 ([95]). Assume that the first order variational equation VE 1 splits
into a direct sum of Lamé–type equations with n1 , n2 integers (A.8). Then Gk is
commutative if and only if the solutions of VE k are meromorphic functions with
respect to variable t.
Proof. The proof is easy. The monodromy group of each of the VE k is a linear rep-
resentation of the fundamental group of Γ = Γ − {∞} (the point ∞ is represented
in the Weierstrass parametrisation by the origin modulo periods) and this funda-
mental group is free, non-commutative and generated by the translations along the
periods. The commutator of these two generators is represented by a simple loop
around the singular point ∞. Hence, a monodromy group is commutative if and
only if the monodromy associated to this simple loop is trivial. By Zariski clo-
sure, a differential Galois group Gk is commutative if and only if the corresponding
monodromy subgroup is commutative.
Therefore we can check the commutativity of Gk locally at ∞. Recursively, by
local power series expansions of the solutions of VE k−1 and quadratures, it is easy
to check whether VE k has branched solutions around 0. One only needs to check
208
206
for the existence of a residue different from zero, which will give rise by integration,
when we apply the method of variations of constants, to a local logarithm.
Assume both equations in (A.8) fall in the Lamé case with a particular solution
in the coefficient field of elliptic functions (see Appendix C); then the Galois group
of (A.8) is given by unipotent matrices of the type
⎛ ⎞
1 0 0 0
⎜ α 1 0 0 ⎟
(A.9) ⎜ ⎟,
⎝ 0 0 1 0 ⎠
0 0 β 1
and G1 ⊂ (C2 , +). Necessarily G1 is trivial or either (C, +) or (C2 , +). In any case
G1 = (G1 )0 is commutative. Hence, by Lemma A.3 Gk = (Gk )0 and by Lemma
A.4 (Gk )0 is commutative if, and only if, the solutions of VE k are meromorphic
functions in the variable t.
In fact we can go further in our analysis. The following result is new.
Lemma A.5. Assume that the first order variational equation VE 1 splits into
a direct sum of Lamé–type equations with n1 , n2 integers (A.8) and that the Ga-
lois group of the first order normal variational equation is not finite. Then Gk is
commutative if and only if (Gk )0 is commutative.
Proof. We first study the Galois group G1 of the first order variational equations.
The tangential variational equation has a solution in the field of meromorphic func-
tions over the elliptic integral curve Γ, K = M(Γ), it falls into the Lamé case
and its Galois group is connected (Appendix C). Hence, we reduce the problem
to studying the normal first order variational equation NVE, for instance the first
equation in (A.8),

ξ¨1 = n1 (n1 + 1) ℘(t) + B1 ξ1 .
This equation falls in either
a) the Lamé or
b) Hermite case.
In the first case, one of the particular solutions is a Lamé function ξ1 either
belonging to K or to a quadratic extension K of K, the other independent solution
being transcendent, see Appendix C. If ξ1 ∈ K, then G1 = G01 , therefore Gk = G0k
and the result is trivial. Therefore we can assume ξ1 ∈ K. We only have to prove
that if (Gk )0 is commutative then Gk is also commutative, the converse being
evident.
We assume the contrary: there exists k ∈ N, k ≥ 2, such that G0k is commu-
tative and such that Gk is not commutative – we can assume k minimal: Gk is
commutative for all 1 ≤ k < k. Then, by Lemma A.4, for every 1 ≤ k < k, the
solutions of V Ek are meromorphic functions in the variable t.
K is the field of meromorphic functions of the elliptic curve E = C/(2Zω1 ⊕
2Zω3 ). We consider, see Appendix C, the field K1 of meromorphic functions of the
elliptic curve E1 = C/(4Zω1 ⊕4Zω3 ). The identity of C induces a map π : E1 → E
and an inclusion of fields K ⊂ K1 , π is a covering of order 4 and Gal(K1 /K) is
a group of order 4 isomorphic to Z2 ⊕ Z2 . We have differential field inclusions
K ⊂ K ⊂ K1 and Gal(Lk /K) = G0k , therefore the action of the monodromy of E1
on the solutions of V Ek is abelian.
209
67
We consider the parallelogram P ⊂ C defined by the four points −ω1 −

ω3 , 3ω1 − ω1 , 3ω1 + 3ω3 , −ω1 + 3ω3 . It is a fundamental domain for the elliptic
curve E1 = C/(4Zω1 ⊕ 4Zω3 ) and the oriented boundary ∂P of P corresponds to
the commutator of two fundamental loops of E1 .
If Gk is not commutative, then there exists a solution of V Ek with a ramification
at zero (mod (ω1 , ω3 )), which is the unique singularity of V Ek (mod. (2ω1 , 2ω3 ));
therefore, if we compute the solutions of V Ek using expressions Σ in the solutions
of the V Ek , for 1 ≤ k < k, and a quadrature, we see that one of the scalar
components ζ of one of the expressions Σ must be a meromorphic function in the
variable t, with a pole at 0 such that the corresponding residue a is not trivial (if
it is not the case, then all the solutions of V Ek in the variable t are meromorphic
and Gk is commutative). If we interpret ζ as a (perhaps ramified) function on E1 ,
i.e. mod. (4ω1 , 4ω3 ), we get exactly four poles, at 0, 2ω1 , 2ω2 , 2ω1 + 2ω2 (using
the variable t, there are 4 poles in the fundamental domain P), the corresponding
residues of ζ being respectively a, a, a, a (the respective singularities correspond by
translations). Then the action of the loop ∂P on ζ(t)dt is ζ(t)dt + 8aiπ and is
not trivial. Therefore the action of the commutator of two fundamental loops of
E1 on some particular solution of V Ek is not trivial. This is a contradiction.
Assume we are in case b). The proof is now easier, because the Galois group
of the NVE must then be the multiplicative group, since we assume that its Galois
group is not finite. Hence, the Galois group G1 is connected, the higher order Galois
group Gk being also connected: Gk = (Gk )0 .
We observe that in the proof we used the fact that the tangential variational
equation always falls in the Lamé case, one of its solutions being in the field K of
elliptic functions: one particular solution is given by the temporal derivative of the
function which defines the integral curve Γ. On the other hand, we know that in
the Lamé case the Galois group is not finite (see Appendix C). Then, as a corollary
of the above Lemmas and of Theorem 2.10, we obtained the following logarithmic
non-integrability criterium.
Proposition A.6. Assume that:
(1) The first order variational equation VE 1 splits in a direct sum of Lamé
type equations, (A.8), with n1 , n2 integers (the tangential and the normal
variational equations, NVE),
(2) the Galois group of the NVE is not finite.
Then a sufficient condition for meromorphic non-integrability is the existence of

a local logarithm at t = 0 in a particular solution of a higher order variational
equation VE k , k > 1.
We remark that in all applications of this proposition considered in this con-

tribution (for example, in Section 2.4) we have a two-degree-of-freedom Hamilton-
ian system with a potential V = V (x1 , x2 ) and an invariant plane (for instance,
x2 = y2 = 0). Then the integral curve is defined by x2 = y2 = 0, x1 = x1 (t) ∈ K,
y1 = y1 (t) = ẋ1 (t) ∈ K, and y1 (t) is a particular solution of the tangential varia-
tional equation. Furthermore, the above proposition is currently the only practical
criterion being used in connection with our general higher order non-integrability
theorem (Theorem 2.10) and specifically exploiting V Ek for k > 1.
210
208
b) Boucher–Weil criterion. In their studies on the integrability of the Three-

Body Problem, Boucher and Weil introduced the following criterion for the non–
commutativity of the Galois group. If the NVE is a system of dimension 2(n − m),
by the cyclic vector method, it is possible to obtain a scalar linear differential
equation of order 2(n − m),
(A.10) L(ξ) = 0,
being L a linear differential operator of order 2(n − m) ([139]). In fact, the elim-
ination process for obtaining equation (6.6) from (6.5) in section 6 is a particular
case of this method. Then using Theorem 2.7, and their own result about the non–
commutativity of the identity component of the Galois group for equation (A.10)
in presence of logarithmic terms, they obtained the following theorem.
Theorem A.7 ([16, 17, 19]). If equation (6.6) has a completely reducible
factor whose local solutions at a singular point contain logarithmic terms, then the
Hamiltonian system XH is not integrable with meromorphic first integrals.
A particular case of the above theorem is for a NVE with only one irreducible
factor. Thus, if equation (6.6) is irreducible and has local solutions at a singular
point with logarithmic terms, then the Hamiltonian system is not integrable.
We illustrate this criterion with an example taken from the study of a FRW
cosmological model in [20]. The Hamiltonian is given by
1 1 Λ λ
(A.11) H = (y12 + y22 ) + x22 + x21 − m2 x21 x22 + x41 + x41 ,
2 2 2 2
obtained from the Hamiltonian HF RW of section 5 for k = 1 and with the usual
change (x1 , y1 ) → (−ix1 , iy1 ).
As was said in Section 5.2, this Hamiltonian has the invariant plane x1 =
y1 = 0 which, on the energy level h = 0, defines a particular solution with NVE ;
furthermore, by means of the algebrization procedure of Section A.2, an algebraic
form is obtained for the NVE,
2 2 2m2 2m2
(A.12) (3x − 1)(3x + 1)2 ξ + (3x + 1)(3x − 1)ξ + x+1+ ξ=0
3 3 λ 3λ
This equation is Fuchsian with three (regular) singular points: x = −1/3, 2/3
and ∞ (it can be reduced to an hypergeometric equation, but we do not use this
fact here).
The exponents at the singular point x = −1/3 are 1/2 and −1/2 (roots of the
indicial equation at that point). Then the difference of exponents is an integer,
and if one of the solutions is ξ1 = (3x + 1)1/2 f (x), f (x) without singularities at
x = −1/3, the other solution has a logarithmic term provided m is different from
zero.
In order to apply Theorem A.7, we must study the necessary conditions for
reducibility. At the singular point x = 2/3 the exponents are α1 = 0 and α2 = 1/2
and at x = ∞ the indicial equation is
2λρ2 − λρ + m2 = 0.
Using the reference [121] the authors obtained the following. If the equation (A.12)
is reducible it must have an exponential solution of the type
211
69
(A.13) ξ = (3x + 1)1/2 (3x − 2)αi P (x),

where αi is one of the exponents at x = 2/3 (i.e., either 0 or 1/2), P (x) is a
polynomial of degree d and
1
d + + αi + ρi = 0,
2
ρi being one of the roots of the indicial equation at ∞.
Then a necessary condition for the existence of the exponential solution (A.13)
is obtained, namely that ρi be equal to either −d − 1 or −d − 1/2. And this implies
2m2
λ=− ,
(p + 1)(p + 2)
with p ∈ N. By an obvious symmetry argument, an analogous condition is ob-
tained for the other parameter Λ. In this manner, Boucher and Weil recovered the
necessary integrability conditions obtained in Coelho, Skea and Stuchi in [28] (see
Section 5.2).
In a similar way it is possible to study the commutativity of the identity com-
ponent of the Galois group in presence of irregular singular points; see the example
in [17], pag. 98 , where this method is applied to a Hénon-Heiles type Hamiltonian
system with a NVE of Bessel type, studied in [97, 95] by means of other methods.
Appendix B. Hypergeometric Equation

The hypergeometric (or Riemann) equation is the most general second order
linear differential equation over the Riemann sphere with three regular singular
singularities. If we place the singularities at x = 0, 1, ∞ it is given by
d2 ξ 1 − α − α 1 − γ − γ dξ
+ ( + )
dx2 x x−1 dx
αα γγ ββ − αα − γγ
(B.1) + ( 2 + + )ξ = 0,
x (x − 1)2 x(x − 1)
where (α, α ), (γ, γ ),(β, β ) are the exponents at the singular points and must
satisfy the Fuchs relation α + α + γ + γ + β + β = 1. We denote the exponent
differences by λ̂ = α − α , ν̂ = γ − γ and µ̂ = β − β .
We also use one of its reduced forms
d2 ξ c − (a + b + 1)x dξ ab
(B.2) + − ξ = 0,
dx2 x(x − 1) dx x(x − 1)
where a, b, c are parameters, with the exponent differences λ̂ = 1 − c, ν̂ = c − a − b
and µ̂ = b − a, respectively.
Now, we recall a theorem of Kimura giving necessary and sufficient conditions
for the integrability of the hypergeometric equation.
Theorem B.1 ([57]). The identity component of the Galois group of the hy-
pergeometric equation (B.1) is solvable if, and only if, either
(i) at least one of the four numbers λ̂ + µ̂ + ν̂, −λ̂ + µ̂ + ν̂, λ̂ − µ̂ + ν̂, λ̂ + µ̂ − ν̂ is
an odd integer, or
(ii) the numbers λ̂ or −λ̂, µ̂ or −µ̂ and ν̂ or −ν̂ belong (in an arbitrary order) to
one or more of the following fifteen families
212
210
1 1/2 + l 1/2 + m arbitrary complex number

2 1/2 + l 1/3 + m 1/3 + q
3 2/3 + l 1/3 + m 1/3 + q l + m + q even
4 1/2 + l 1/3 + m 1/4 + q
5 2/3 + l 1/4 + m 1/4 + q l + m + q even
6 1/2 + l 1/3 + m 1/5 + q
7 2/5 + l 1/3 + m 1/3 + q l+m+q even
8 2/3 + l 1/5 + m 1/5 + q l+m+q even
9 1/2 + l 2/5 + m 1/5 + q l+m+q even
10 3/5 + l 1/3 + m 1/5 + q l+m+q even
11 2/5 + l 2/5 + m 2/5 + q l+m+q even
12 2/3 + l 1/3 + m 1/5 + q l+m+q even
13 4/5 + l 1/5 + m 1/5 + q l+m+q even
14 1/2 + l 2/5 + m 1/3 + q l+m+q even
15 3/5 + l 2/5 + m 1/3 + q l+m+q even
Here l, m and q are integers.
We recall that Schwarz’s table provides us with the cases for which the Galois
(and monodromy) groups are finite (i.e., the identity component of the Galois group
is reduced to the identity element) and is given by fifteen families. These are given
by families 2–15 of the table above and by the family (1/2 + Z) × (1/2 + Z) × Q
(see, for instance, [114]). Since the latter family is already contained in family 1
in the above table, so are, of course, all families due to Schwartz.
Appendix C. Lamé Equation

The algebraic form of the Lamé Equation is [114, 141]
d2 ξ f (x) dη Ax + B
(C.1) 2
+ − ξ = 0,
dx 2f (x) dx f (x)
where f (x) = 4x3 − g2 x − g3 , with A, B, g2 and g3 parameters such that the
discriminant of f , 27g32 − g23 is non-zero. This equation is a Fuchsian differential
equation with four singular points over the Riemann sphere.
With the well–known change x = ℘(t), we get the Weierstrass form of the Lamé
equation
d2 ξ
(C.2) − (A℘(t) + B)ξ = 0,
dt2
where ℘ is the elliptic Weierstrass function with invariants g2 , g3 (we recall that ℘(z)
is a solution of the differential equation ( dx 2
dt ) = f (x)). It is a 4-parametric family
of equations in the parameters A, B, g2 and g3 . Classically the equation is written
with the parameter n instead of A, with A = n(n + 1). This equation is defined
on a torus Π (a genus one Riemann surface or elliptic curve y 2 = f (x)) with only
one singular point at the origin. It is also a Fuchsian linear differential equation.
Let 2ω1 , 2ω3 be the two periods of the Weierstrass function ℘ and g1 , g2 their
corresponding monodromies along these periods. If g∗ represents the monodromy
around the singular point, then g∗ = [g1 , g2 ] ([141, 114]). The Lamé equation in
the form (C.2) was intensively studied by Halphen [42].
213
71
By Theorem A.2 of Appendix A we know that the identity component of the

Galois group is preserved by the change of variables Π → P1 , t → x. The relation
between the monodromy groups of equations (C.1) and (C.2) is discussed in [114],
section IX and, from a modern point of view, in [27].
Now the known mutually exclusive cases of closed form solutions of the Lamé
equation (C.2) are as follows:
(i) The Lamé–Hermite case [36, 42, 114, 141]. In this case n ∈ Z the three
other parameters are arbitrary.
(ii) The Brioschi-Halphen-Crawford solutions [13, 36, 42, 114]. Now m :=
2 ∈ N and the parameters B, g2 and g3 must satify an algebraic equation
1
n+
0 = Qm (g2 /4, g3 /4, B) ∈ Z[g2 /4, g3 /4, B],

where Qm has degree m in B. This polynomial is known as the Brioschi determi-
nant.
(iii) The Baldassarri solutions [13]. The condition on n is n + 12 ∈ 13 Z ∪ 14 Z ∪
1
5Z − Z, with additional (involved) algebraic restrictions on the other parameters.
It is possible to prove that the only integrable cases of integrability for the Lamé
equation are the cases (i)–(iii) above. Integrability here means in the Galois sense,
where the coefficient field is the field of elliptic functions C(℘(t), ℘ (t)), isomorphic
to the field of meromorphic functions on the torus Π.
Proposition C.1 ([93],[95]). Equation (C.2) is integrable only in the cases
(i), (ii) and (iii) above.
Corollary C.2. A necessary condition for the commutativity of the identity
component of the Galois group of equation (C.2) is that the latter belong to one of
the cases (i), (ii) or (iii) above.
We recall that the moduli of the elliptic curve y 2 = 4x3 − g2 x − g3 (we write the
elliptic curve in the canonical form, where as above g2 and g3 are the invariants) is
characterized by the value of the modular function j,
g23
(C.3) j = j(g2 , g3 ) = .
g23 − 27g32
We recall that two elliptic curves are birationally equivalent if, and only if, they
have the same value of the modular function (see, for instance [120]).
Although the conditions on g2 , g3 and B for a finite Galois group (case (iii)) are
difficult to systematize, there is, in this case, a general result by Dwork answering
a question posed by Baldassarri in [13].
Proposition C.3 ([37]). Assume that the Galois group of equation (C.2) is
finite. Then for a fixed value of n, the number of possible couples (j, B) is finite.
We note that the proof by Dwork was stated for the algebraic form of the Lamé
equation (equation (C.1)). But since the identity component of the Galois group
is preserved by a finite covering (Theorem A.1 of Appendix A), then the finiteness
of the Galois group of equation (C.1) is equivalent to the finiteness of the Galois
214
212
group of equation (C.2) (a linear algebraic group is finite if, and only if, its identity
component is trivial) and the result is valid also for equation (C.2).
The first author is indebted to B. Dwork for sending the above result.
One more reference about the case (iii) of Lamé equation is [82]. This reference
corrected a mistake in the paper [13].
Now we center our attention on the classical Lamé–Hermite case (i). It is easy
to see that a necessary and sufficient condition for the total Galois group of (C.2)
to be commutative is that n ∈ Z. We sketch the steps of the proof. Indeed, this is
a classical well-known necessary and sufficient condition for the monodromy group
M of the equation (C.2) to be commutative (it is clear that, as M is generated by
g1 and g2 , an equivalent condition for the commutativity of M is g∗ = 1 (identity),
and the indicial equation at the singularity is ρ2 − ρ − n(n + 1) = 0, and there is
no logarithmic term for integer n (see [114]). Therefore, since G is topologically
generated by M , it must also be commutative.
There are two excluding cases for (i):
(1) (Lamé) There is one solution which is a Lamé function ξ1 either belong-
ing to the coefficient field K = C(℘(t), ℘ (t)) or such that ξ12 belongs to
K ([114]). Hence one solution belongs to a quadratic extension of the
coefficient field. The other independent solution ξ2 is transcendent over
the field K ([13]). For a fixed n ∈ Z, the parameters B, g2 and g3 must
satisfy also an algebraic equation, 0 = P2n+1 (g2 , g3 , B) ∈ Q[g2 , g3 , B], of
degree 2n + 1 in B ([42]).
(2) (Hermite) There are two particular C-independent solutions ξ1 , ξ2 , such
that the product ξ1 ξ2 belong to the field K. No other conditions are
satisfied for the parameters, except that n ∈ Z and they do not satisfy
the algebraic conditions of the Lamé case ([42, 114, 141].
In the case of Lamé when ξ1 ∈ K, the Galois group G of equation (C.2) is
connected of the type 2 of Proposition 2.5 in Section 2.1:

1 0
G = G0 = ,µ∈C .
µ 1
When ξ12 ∈ K, ξ1 ∈ / K, we have 2ξ1 ξ1 ∈ K, therefore ξ1 ∈ K. We set K =
K(ξ1 ), then Gal(K/K) is a cyclic group of order 2 and K = K(ξ1 , ξ1 ). Setting
L = K(ξ1 , ξ2 , ξ1 , ξ2 ) = K(ξ2 , ξ2 ), then Gal(L/K) is connected and isomorphic to
the additive group C.
By the Galoisian correspondence (see Section 2.2), the algebraic closure of K,
in the Picard-Vessiot extension L = K(ξ1 , ξ2 , ξ1 , ξ2 ) is K and we have the chain of
differential fields
K ⊂ K ⊂ L,
0 0
with G/G = Gal(K/K) and G = Gal(L/K).
In this case, the Galois group G of equation (C.2) is of the type 3 of Proposition
2.5 of Section 2.1:

λ 0
G = G2 = , λ is a 2-root of unity, µ ∈ C ,
µ λ−1

1 0
0
G = ,µ∈C .
µ 1
215
73
The extension K ⊂ L is a purely transcendent Picard-Vessiot extension with

an associated linear differential equation which can be made explicit by means of
the Halphen transformation [42, 114].
The field K is the field of meromorphic functions of the elliptic curve E =
C/(2Zω1 ⊕2Zω3 ). We consider the field K1 of meromorphic functions of the elliptic
curve E1 = C/(4Zω1 ⊕ 4Zω3 ). The identity of C induces a map π : E1 → E and
an inclusion of fields K ⊂ K1 , π is a covering of order 4 and Gal(K1 /K) isomorphic
to Z2 ⊕ Z2 . We can interpret K as a subfield of K1 (see [114]). First we perform
the change of independent variable t = 2τ (which induces an isomorphism between
the elliptic curves E1 and E) and use the addition theorem for ℘ (see [114]) we
obtain
+ 1 ℘ (τ ) 2 ,
d2 ξ
(C.4) − 4 n(n + 1) − 2℘(τ ) + B ξ = 0.
d τ2 4 ℘ (τ )
Now, in order to complete the Halphen transformation, we perform the change
−n
ξ = ℘ (τ ) η, obtaining
d2 η ℘ (τ ) dη
(C.5) 2
− 2n
+ 4 n(2n − 1) ℘(τ ) − B η = 0 ,
dτ ℘ (τ ) dτ
with singularities at τ = 0 (modulo the periods (2ω1 , 2ω3 )). In other words, equa-
tion (C.5) corresponds to the Picard-Vessiot extension K ⊂ L with a connected
Galois group G0 and, just like Lamé’s equation, it is also a Fuchsian linear differ-
ential equation defined over an elliptic curve (a copy of the initial one) with only
one singular point.
For some other applications of the Lamé equation, different from those consid-
ered in this contribution, see [95], sections 6 and 7.
References
[1] M. Abramowitz, I.A. Stegun Editors, I. A. Handbook of mathematical functions with formulas,
graphs, and mathematical tables, John Wiley & Sons Inc., New York, 1984.
[2] P. Acosta-Humánez, Nonautonomous Hamiltonian Systems and Morales-Ramis Theory I. The
Case ẍ = f (x, t), SIAM J. Appl. Dyn. Syst. 8 (2009), 279–297.
[3] P. Acosta-Humánez, D. Blázquez-Sanz, Non-integrability of some Hamiltonians with rational
potentials, Discrete Contin. Dyn. Syst. 10 (2008), 265–293.
[4] A. Almeida, A. López-Castillo, T. Stuchi, Non-integrability proof of the frozen planetary atom
configuration, J. Phys. A: Math. Gen. 36 (2003) 4805–4814.
[5] A. Almeida, T.J. Stuchi, The integrability of the anisotropic Stormer problem with angular
momentum, Physica D 189 (2004) 219–233.
[6] V.I. Arnold, Mathematical methods in classical mechanics. Springer-Verlag, Berlin, 1978.
[7] M. Arribas, A. Elipe, Non-integrability of the motion of a particle around a massive straight
segment, Physics Letters A 281 (2001) 142–148.
[8] M. Arribas, A. Elipe, A. Riaguas, Non-integrability of anisotropic quasihomogeneous Hamil-
tonian systems, Mech. Res. Comm. 30 (2003) 209–216.
[9] M. Audin, Les systèmes Hamiltoniens et leur intégrabilité, Cours Spécialisés, Collection SMF
8 Société Mathematique de France, Marseille 2001.
[10] M. Audin, La réduction symplectique appliquée à la non-intégrabilité du problème du satellite,
Ann. Fac. Sci. Toulouse Math. 12 (2003) 25–46.
[11] M. Audin Exemples de hamiltoniens non intégrables en mécanique analytique réelle, Ann.
Fac. Sci. Toulouse Math. 12 (2003), 1-23
[12] M. Audin, Mon choix de Sophie, preprint 2006 (http://www-irma.u-
strasbg.fr/ maudin/choix-sophie.pdf).
216
214
[13] F. Baldassarri, On algebraic solutions of Lamé’s differential equation, J. of Diff. Eq. 41,
(1981), 44-58.
[14] B. S. Bardin, A. J. Maciejewski, A. J., M. Przybylska, Integrability of generalized Jacobi
problem, Regul. Chaotic Dyn. 10 (2005), 437–461.
[15] F. Beukers, Differential Galois Theory, From Number Theory to Physics, W. Waldschmidt,
P. Moussa, J.-M. Luck, C. Itzykson Ed., Springer-Verlag, Berlin 1995, 413–439.
[16] D. Boucher, Sur la non-intégrabilité du problème plan des trois corps de masses égales, C.R.
Acad. Sci. Paris Série I 331 (2000) 391–394.
[17] D. Boucher, Sur les équations différentielles paramétrées, une application aux systèmes hamil-
toniens, Thèse Faculté des Sciencies de Limoges 2000.
[18] D. Boucher, Non complete integrability of a satellite in circular orbit. Port. Math. (N.S.) 63
(2006) 69–89.
[19] D. Boucher, J.-A. Weil, Application of J.-J. Morales and J.-P. Ramis’ theorem to test the non-
complete integrability of the planar three-body problem. From combinatorics to dynamical
systems, 163–177, IRMA Lect. Math. Theor. Phys., 3, de Gruyter, Berlin, 2003.
[20] D. Boucher, J.-A. Weil, On the non-integrability in the Friedmann-Robertson-Walker Cos-
mological Model Brazilian J. Phys. 37 (2007) 398–405.
[21] A. Borel, Linear algebraic groups, Springer-Verlag, New-York, 1991.
[22] R. Carter, G. Seagal, I. Macdonald, Lectures on Lie Groups and Lie Algebras, Cambridge
University Press, Cambridge, UK, 1995.
[23] G. Casale, Sur le groupoı̈de de Galois d’un feuilletage. Thèse, Toulouse 2004.
[24] G. Casale, Le groupe de Galois de P1 et son irreductibilité, Comment. Math. Helv. 83 (2008)
471–519.
[25] G. Casale, The Galois groupoid of Picard Painlevé VI equation, in Algebraic, analytic and
geometric aspects of complex differential equations and their deformations, Painlevé hierar-
chies RIMS Kôkyûroku Bessatsu, B2 (2007), 15–20.
[26] R.C. Churchill, J. Delgado, D.L. Rod, M. Alvarez, J. Delgado, The spring-pendulum system
and the Riemann equation, New trends for Hamiltonian systems and celestial mechanics
(Cocoyoc, 1994), 97–103, Adv. Ser. Nonlinear Dynam. 8, World Sci. Publ., River Edge, NJ,
1996
[27] R.C. Churchill, Two Generator Subgroups of SL(2, C) and the Hypergeometric, Riemann
and Lamé Equations, J. Symbolic Computation 28 (1999) 521–545.
[28] L.A.A. Coelho, J.E.F. Skea, T.J. Stuchi, On the Non-integrability of a Class of Hamiltonian
Cosmological Models, Brazilian J. Phys. 35 (2005) 1048–1049.
[29] L.A.A. Coelho, J.E.F. Skea, T.J. Stuchi, Friedmann Robertson Walker models with Confor-
mally Coupled Massive Scalar Fields are Non-integrable, preprint 2006.
[30] E. Compoint, M.F. Singer, Computing Galois groups of Completely Reducible Differential
Equations, J. Symbolic Computation 11 (1998) 1–22.
[31] J. Drach, Essai sur une théorie générale de l’intégration et sur la classification des transcen-
dantes Ann. Sci. École Normale Sup. 15 (1898) 243–324.
[32] H. Dullin, A. Tsygvintsev, On the analytic non-integrability of the Rattleback problem,
Annales de la faculté des sciences de Toulouse, 17 (2008)495-517.
[33] A. Duval, M. Loday-Richaud, Kovacic’s algorithm and its application to some families of spe-
cial functions, Applicable Algebra in Engineering, Communication and Computing 3 (1992),
211-246.
[34] A. Duval, The Kovacic Algorithm with applications to special functions. Differential Equa-
tions and Computer Algebra, M. Singer, Ed., Academic Press, London, 1991, 113-130.
[35] P. Du Val, Elliptic functions and elliptic curves, Cambridge University Press, 1973.
[36] B. Dwork, Differential operators with nilpotent p-curvature, Am. J. Math. 112 (1990), 749-
786.
[37] B. Dwork, private communication.
[38] S. Ferrer, F. Mondéjar, On the non-integrability of the Stark–Zeemann Hamiltonian system,
Comm. in Math. Phys. 208 (1999) 55–63.
[39] S. Ferrer, F. Mondéjar, On the non-integrability of the 3-D hydrogen atom under motional
Stark effect or circularly polarized microwave combined with magnetic fields, Phys. Lett. A
264 (1999) 74–83.
[40] S. Ferrer, F. Mondéjar, On the non-integrability of the Generalized van der Waals Hamilton-
ian, J. of Math. Phys. 41 (2000) 5445–5452.
217
75
[41] V.I. Gromak, I. Laine, S. Shimomura, Painlevé Differential Equations in the Complex Plane,
Walter de Gruyter, Berlin 2002.
[42] G. H. Halphen, Traité des fonctions elliptiques Vol. I, II. Gauthier-Villars, Paris, 1888.
[43] J. Hietarinta, A search for integrable two-dimensional Hamiltonian systems with polynomial
potential, Physics Letters A 96 (1983) 273–278.
[44] J. Hietarinta, Direct methods for the search of the second invariant Phys. Rep. 147 (1987)
87–154.
[45] G. W. Hill, Researches in the Lunar theory, American Journal of Mathematics 1 (1878), p.
5-6, 129-147, 245-260.
[46] E. Horozov,T. Stoyanova, Non-Integrability of Some Painlevé VI-Equations and Dilogarithms
Regular and chaotic Dynamics, 12 (2007) 622-629.
[47] J.E. Humphreys, Linear Algebraic Groups, Springer-Verlag, New York, 1981.
[48] E. L. Ince, Ordinary differential equations, Dover, Nova York, 1956.
[49] D.I. Goryachev, New integrable cases of integrability of the Euler dynamical equations, War-
saw Univ. Izv. 1910 (in russian).
[50] H. Ito, Non-integrability of the Hénon-Heiles system and a theorem of Ziglin, Koday Math.
J. 8 (1985) 129-138.
[51] K. Iwasaki, Some dynamical aspaects of Painlevé VI Algebraic Analysis of Differential Equa-
tions in honor of Prof. Takahiro Kawai on the occasion of his sixtieth birthday, T. Aoki, Y.
Takei, N. Tose, H. Majima eds.(2007) 143-156.
[52] E. Juillard Tosel, Un rèsultat de non-integrabilité pour le potentiel en 1/r 2 , C. R. Acad. Sci.
Paris, t. 327, Série I (1998) 387–329.
[53] E. Juillard,Non-integrabilité algebrique et méromorphe de problèmes de N corps, Thése Univ.
Paris VI, Janvier 1999.
[54] E. Juillard Tosel, Un nouveau critère de non-integrabilité méromorphe d’un Hamiltonian, C.
R. Acad. Sci. Paris, t. 330, Série I (2000) 1097–1102.
[55] I. Kaplansky, An Introduction to Differential Algebra. Hermann, Paris 1976.
[56] N.M. Katz, A conjecture in the arithmetic theory of differential equations. Bull. Soc. Math.
France 110 (1982), 203-239.
[57] T. Kimura, On Riemann’s Equations wich are Solvable by Quadratures, Funkcialsj Ekvacioj
12 (1969) 269–281.
[58] E. Kolchin, Differential algebra and algebraic groups. Academic Press, New York, 1973.
[59] J.J. Kovacic, An Algorithm for Solving Second Order Linear Homogeneous Differential Equa-
tions. J. Symbolic Computation 2 (1986), 3-43.
[60] S. Kowalevski, Sur le probleme de la rotation d’un corps solide autour d’un point fixe, Acta
Math. 12 (1889), 177-232.
[61] V.V. Kozlov, Non-existence of univalued integrals and branching of solutions in rigid body
dynamics Pikl. Mat. Mekh. 42, no. 3 (1978), 400-406.
[62] M. Kummer, A.W. Saenz, Nonintegrability of the Zeeman Hamiltonian, Commun. in Math.
Phys. 162 (1994) 447–465.
[63] M. Kummer, A.W. Saenz, Nonintegrability of the Stormer problem, Physica D 86 (1995)
363–372.
[64] J.L. Lagrange, Ouvres, Vol. 6, Paris 1873.
[65] L. Landau, E. Lifchitz, Théorie des champs. Mir, Moscou 1970.
[66] F. Loray, S. Cantat, Holomorphic Dynamics, Painlevé VI equation and character varieties,
preprint, Université de Rennes, 2007.
[67] A.J. Maciejewski, M. Szydlowski, Towards a description of complexity of the simplest cos-
mological systems, J. Phys.A: Math. Gen. 33 (2000) 9241–9254.
[68] A.J. Maciejewski, M. Szydlowski, Integrability and Non-integrability of Planar Hamiltonian
Systems of Cosmological Origin, J. of Nonlinear Math. Phys. 8 (2001) 200–206.
[69] A.J. Maciejewski, M. Przybylska. Non-integrability of restricted two-body problems in con-
stant curvature spaces. Regul. Chaotic Dyn., 8 (4) (2003) 413–430.
[70] A.J. Maciejewski, M. Przybylska, Non-integrability of the problem of a rigid satellite in
gravitational and magnetic fields, Celestial Mech. Dynam. Astronom. 87 (2003) 317–351.
[71] A.J. Maciejewski, M. Przybylska, Non-integrability of the Suslov problem, J. Math. Phys. 45
(2004) 1065–1078.
[72] A.J. Maciejewski, M. Przybylska, All Meromorphically Integrable 2D Hamiltonian Systems
with Homogeneous Potential of Degree 3, Phys. Lett. A 327 (2004) 461–473.
218
216
[73] A.J. Maciejewski, M. Przybylska, Non-integrability of the generalized two fixed centres prob-
lem, Celestial Mech. Dynam. Astronom., 89 (2004), 145–164.
[74] A.J. Maciejewski, M. Przybylska, Differential Galois Approach to the Non-integrability of
the Heavy Top Problem, Annales de la Faculté des Sciences de Toulouse Sér. 6, 14 (2005)
123–160.
[75] A.J. Maciejewski, M. Przybylska, Darboux points and integrability of Hamiltonian systems
with homogeneous polynomial potential, J. Math. Phys. 46 (2005) 062901.1–062901.33.
[76] A.J. Maciejewski, M. Przybylska, T. Stachowiak, Non integrability of Gross-Neveu systems,
Physica D 201 (2005) 249–267.
[77] A.J. Maciejewski, M. Przybylska, T. Stachowiak, M. Szydlowski, Global Dynamics of cosmo-
logical scalar fields-Part I, preprint 2007.
[78] A.J. Maciejewski, M. Przybylska, J.-A. Weil, Non-integrability of the generalized spring-
pendulum problem, J. Phys. A: Math. Gen. 37 (2004) 2579–2597.
[79] A.J. Maciejewski, M. Przybylska, H. Yoshida, Necessary conditions for partial and superin-
tegrability of Hamiltonian systems with homogeneous potential, preprint 2007.
[80] A. J. Maciejewski, J. M. Strelcyn, M. Szydlowski, Non-integrability of Bianchi VIII Hamil-
tonian System, J. Math. Phys.42 (2001), 1728–1743.
[81] A.J. Maciejewski, W. Respondek, The nilpotent tangent 3-dimensional sub-Riemannian prob-
lem is nonintegrable, Decision and Control 2004. CDC. 43rd IEEE Conference on, 438- 443
Vol.1.
[82] R. Maier, Algebraic Solutions of the Lamé Equation, Revisited, J. of Diff. Equations 198
(2004), 16-34.
[83] B. Malgrange, Le groupoı̈de de Galois d’un feuilletage, L’enseignement mathématique 38,
vol 2 (2001) 465–501.
[84] B. Malgrange, On the non linear Galois theory, Chinese Ann. Math. Ser. B 23, 2 (2002)
219–226.
[85] J. Malmquist, Sur les équations différentielles du second ordre dont l’intégrale générale a ses
points critiques fixes, Arkiv. Mat., Astron. Fys. 18(8)(1922) 1-89.
[86] J. Martinet, J.P. Ramis, Théorie de Galois différentielle et resommation.Computer Algebra
and Differential Equations, E. Tournier ed. Academic Press, London, 1989, 117–214.
[87] R. Martı́nez, C. Simó, Non-Integrability of Hamiltonian Systems Through High Order Varia-
tional Equations: Summary of Results and Examples, Regular and Chaotic Dynamics (2009)
14 (3) 323-348.
[88] R. Martı́nez, C. Simó, Non-Integrability of the degenerate cases of the Swinging Atwood’s
Machine using higher order variational equations, preprint (2009).
[89] R. Martı́nez, C. Simó, Efficient numerical implementation of integrability criteria based on
high order variational equations, preprint (2009).
[90] Richard Moeckel, On central configuration, Math. Z. 205 (1990), 499-517.
[91] F. Mondéjar, Non-integrability of parametric Hamiltonian systems by differential Galois the-
ory, Monografı́as de la Acad. de Ciencias. Zaragoza 14 (1999) 60-66.
[92] F. Mondéjar, On the Non-integrability of Hamiltonian Systems with sum of Homogeneous
Potentials, unpublished.
[93] J.J. Morales-Ruiz, C. Simó, Picard-Vessiot Theory and Ziglin’s Theorem. J. of Diff. Equations
107 (1994), 140-162.
[94] J.J. Morales-Ruiz, C. Simó, Non-integrability criteria for Hamiltonians in the case of Lamé
Normal Variational Equations. J. of Diff. Equations 129 (1996) 111–135.
[95] J. J. Morales-Ruiz, Differential Galois Theory and Non-Integrability of Hamiltonian Systems.
Birkhäuser, Basel 1999.
[96] J.J. Morales-Ruiz, J.M. Peris, On a Galoisian Approach to the Splitting of Separatrices, Ann.
Faculté Sciencies de Toulouse VIII (1999) 125–141.
[97] J.J. Morales-Ruiz, J.P. Ramis, Galoisian obstructions to integrability of Hamiltonian systems,
Methods and Applications of Analysis 8 (2001) 33–96.
[98] J.J. Morales-Ruiz, J.P. Ramis, A Note on the Non-Integrability of some Hamiltonian Systems
with a Homogeneous Potential, Methods and Applications of Analysis 8 (2001) 113–120.
[99] J.J. Morales-Ruiz, J.P. Ramis, Galoisian Obstructions to integrability of Hamiltonian Systems
II, Methods and Applications of Analysis 8 (2001) 97–102.
[100] J.J. Morales-Ruiz, C. Simó, S. Simon, Algebraic proof of the non-integrability of Hill’s
problem, Ergod. Th. & Dynam. Sys. 25 (2005), 1237–1256.
219
77
[101] J. J. Morales-Ruiz, A Remark about the Painlevé Transcendents, in “Théories asymptotiques

et équations de Painlevé”, S.M.F., Séminaires et Congrès 14 (2005), 229–235.
[102] J.J. Morales-Ruiz, J.P. Ramis, C. Simó, Integrability of Hamiltonian Systems and Differen-
tial Galois Groups of Higher Variational Equations, to appear in Ann. Sc. École Norm. Sup.
40 (2007) 845-884.
[103] J. J. Morales-Ruiz, S. Simon, On the meromorphic non-integrability of some N -body prob-
lems, Discrete and Continuous Dynamical Systems 24 (2009) 1225-1273.
[104] J. Moser, Three integrable Hamiltonian systems , Advances in Math. 16 (1975) 197-220.
[105] K. Nakagawa, Direct construction of polynomial first integrals for Hamiltonian systems
with a two-dimensional homogeneous polynomial potential, Dep. of Astronomical Science,
The Graduate University for Advanced Study and the National Astronomical Observatory of
Japan, Ph. D. Thesis 2002.
[106] K. Nakagawa, H. Yoshida, A necessary condition for the integrability of homogeneous Hamil-
tonian systems with two degrees of freedom, J. Phys. A: Math. Gen. 34 (2001) 2137–2148.
[107] K. Okamoto, Polynomial Hamiltonians associated to Painlevé Equations I, IIProc. Japan
Acad. Ser. A 56 (1980) 264-268, 367-371.
[108] K. Okamoto, Studies of the Painlevé Equations, III. Second and Fourth Painlevé Equations,
PII and PIV , Math. Ann. 275 (1986) 221-255.
[109] H. Poincaré, Les Méthodes Nouvelles de la Mécanique Céleste, Vol. I. Gauthiers-Villars,
Paris, 1892.
[110] L. M. Perko, E. L. Walter, Regular polygon solutions of the N -body problem, Proceedings
of the A.M.S. 94 (1985), 301-309.
[111] E. Picard, Sur les groupes de transformation des équations différentielles linéaires, C.R.
Acad. Sci. Paris 96 (1883), 1131–1134.
[112] E. Picard, Sur équations différentielles et les groupes algébriques des transformation, Ann.
Fac. Sci. Univ. de Toulouse (1) 1(1887), A1–A15.
[113] E. Picard, Traité d’Analyse, Tome III, Gauthiers-Villars, Paris, 1928.
[114] E.G.C. Poole,Introduction to the theory of Linear Differential Equations. Oxford Univ.
Press, London, 1936.
[115] O. Pujol, J.-P. Perez, S. Simon, J.-P. Ramis, J.-A. Weil, C. Simó, Swinging Atwood’s ma-
chine: experimental and theoretical study, preprint (2009).
[116] M. Przybilska, Finitesness of integrable n-dimensional homogeneous polynomial potentials,
Physics Letters A 369 (2007)180-187.
[117] A.W. Sáenz, Nonintegrability of the Dragt-Finn model of magnetic confinament: a
Galoisian-group approach Physica D 144, (2000) 37–43.
[118] S. T. Sadetov, On algebraic integrals of Hill Problem and Restricted Circle Plane Three-
Body Problem on a level of energy, Regul. Chaotic Dyn. 10 323-332 .
[119] A. V. Shchepetilov, Nonintegrability of the two-body problem in constant curvature spaces,J.
Phys. A 39 (2006), 5787–5806.
[120] C.L. Siegel, Topics in complex function theory. Wiley, New York, 1969.
[121] M. Singer, F. Ulmer, Necessary conditions for Liouvillian solutions of (third order) linear
differential equations, Appl. Algebra Engrg. Comm. Comput. 6, (1995).
[122] T. Stuchi, A. López-Castillo,M.A. Almeida, Nonintegrability of the three-body problems for
the classical helium atom, J. Math. Phys. 47 (2006), 093506–093513.
[123] T. J. Stuchi, A. López-Castillo,M. A. Almeida, Nonintegrability of the three-body problems
for the classical helium atom, J. Math. Phys. 47 (2006) 093506.
[124] C. Simó, T. J. Stuchi, Central stable/unstable manifolds and the destruction of KAM tori
in the Planar Hill Problem, Physica D140 (2000), 1-32.
[125] M.F. Singer, F. Ulmer, Galois Groups of second and third order linear differential equations,
J. Symbolic Computation 16 (1993), 1-36.
[126] M.F. Singer, F. Ulmer, Liouvillian and algebraic solutions of second and third order linear
differential equations, J. Symbolic Computation 16 (1993), 37-73.
[127] T. Stachowiak, M. Szydlowski, A.J. Maciejewski, Nonintegrability of density perturbations
in the Friedmann-Robertson-Walker universe, J. Math. Phys. 47 (2006), 032502-032513.
[128] J. Stoer, J., R. Bulirsch, Introduction to numerical analysis, Springer-Verlag, New York,
2002.
[129] T. Stoyanova, O. Christov, Non-Integrability of the Second Painlevé Equation as a Hamil-
tonian system, Compte Rendu de l’académie bulgare des sciences, 60 (2007) 1.
220
218
[130] A.H.Taub, Empty Space-Times admitting a Three Parameter Group of motions. Annals of
Mathematics 53 (1951), 472–490.
[131] A. Tsygvintsev, La non-integrabilité méromorphe du probléme plan des trois corps, C.R.
Acad. Sci. Paris Série I 331 (2000) 241-244.
[132] A. Tsygvintsev, Sur l’absence d’une intégral première méromorphe dans le probléme plan
des trois corps, C.R. Acad. Sci. Paris Série I 333 (2001) 125–128.
[133] A. Tsygvintsev, Sur l’absence d’une intégrale première méromorphe supplémentaire dans le
problème plan des trois corps, C.R. Acad. Sci. Paris, t. 333, Série I (2001) 241–244.
[134] A. Tsygvintsev, The meromorphic non-integrability of the three-body problem, J. Reine
Angew. Math. 537 (2001), 127–149.
[135] A. Tsygvintsev, V. Non-existence of new meromorphic first integrals in the planar three-
body problem, Celestial Mech. Dynam. Astronom. 86 (2003), 237–247.
[136] A. Tsygvintsev, On some exceptional cases in the integrability of the three-body problem,
Celestial Mechanics and Dynamical Astronomy 99 (2007) 23-29.
[137] F. Ulmer, J.A. Weil, Note on Kovacic’s Algorithm, J. Symbolic Computation 22 (1996),
179-200.
[138] H. Umemura, Differential Galois Theory of infinite dimension, Nagoya Mathematical Journal
144 (1996) 59–135.
[139] M. van der Put and M. Singer, Galois Theory of Linear Differential Equations, Springer,
Berlin 2003.
[140] M.E. Vessiot, Sur l’intégration des équations différentielles linéaires, Ann. Sci. de l’École
Norm. Sup. (3) 9 (1892), 197–280.
[141] E.T. Whittaker, E.T. Watson, A Course of Modern Analysis. Cambridge Univ. Press, Cam-
bridge, UK, 1969.
[142] A. Wintner, The analytical foundations of celestial mechanics, Princeton University Press,
Princeton, N. J., 1947 (1941).
[143] K. Yagasaki, Galoisian obstructions to integrability and Melnikov criteria for chaos in two-
degree-of-freedom Hamiltonian systems with saddle centres Nonlinearity 16 (2003), 2003-2012.
[144] K. Yagasaki, Nonintegrability of an infinite-degree-of-freedom model for unforced and un-
damped, straight beams, J. of Applied Mechanics 70 (2003), 732-738.
[145] K. Yagasaki,T. Wagenknecht, Detection of symmetric homoclinic orbits to saddle-centres in
reversible systems, Physica D 214 (2006), 169-181.
[146] H. Yoshida, A criterion for the non-existence of an additional integral in Hamiltonian systems
with a homogeneous potential, Physica D 29 (1987) 128–142.
[147] H. Yoshida, Non integrability of the truncated Toda lattice at any order, Commun. in Math.
Phys. 116 (1988), 529–538.
[148] S.L. Ziglin, Branching of solutions and non-existence of first integrals in Hamiltonian me-
chanics I, Funct. Anal. Appl. 16 (1982), 181–189.
[149] S.L. Ziglin, Branching of solutions and non-existence of first integrals in Hamiltonian me-
chanics II. Funct. Anal. Appl. 17 (1983), 6–17.
[150] S.L. Ziglin, On the absence of a real-analytic first integral in some problems of dynamics.
Funct. Anal. Appl. 31 (1997), 3–9.
[151] S.L. Ziglin, On involutive integrals of groups of linear symplectic transformations and natural
mechanical systems with homogeneous potential. Funktsional. Anal. i Prilozhen. 34 (2000),
26–36.
(J.J. Morales-Ruiz) Universidad Politécnica de Madrid

E-mail address: juan.morales-ruiz@upm.es
(J.-P. Ramis) Université Paul Sabatier

E-mail address: ramis@picard.ups-tlse.fr
Volume 509, 2010
Tournaments and parabolic almost complex structures on

flag manifolds
Marlio Paredes and Sofı́a Pinzón
Dedicated to the memory of Professor Jairo Antonio Charris
Abstract. Using tournaments, we study parabolic almost complex structures

on the classical flag manifold. The correspondence of Burstall and Salamon
between tournaments and invariant almost complex structures on flag mani-
folds provides us a necessary condition for an almost complex structure to be
parabolic. In addition, we give a new proof of the theorem, due to Mo and
Negreiros, which shows that all of the parabolic almost complex structures
admit (1, 2)–symplectic metrics.
1. Introduction
Burstall and Salamon showed in [BS] that there is a one to one correspondence
between invariant almost complex structures on flag manifolds and tournaments, in
a natural way a tournament is associated to each almost complex structure. In that
paper the parabolic almost complex structures are defined using only the associated
tournament.
The relation between tournaments and almost complex structures has been
exploited in several papers. In [MN], Mo and Negreiros used this relation to
construct a family of (1, 2)–symplectic metrics on a flag manifold. They also found
a condition on the associated tournament so that the almost complex structure does
not admit (1, 2)–symplectic metric. Such condition was studied by Paredes in [P2],
[P3], [P4] and was used to produce several examples of (1, 2)–symplectic metrics
on a maximal flag manifold. The following theorem was conjectured by Paredes in
[P2] and it was proved by Cohen, Negreiros and San Martin in [CNS].
Theorem 1.1. The maximal flag manifold (F(n), J), n ≥ 4, admits an in-
variant (1, 2)–symplectic metric if and only if the associated tournament T (J) is
cone–free.
2000 Mathematics Subject Classification. Primary 53C15; Secondary 14M15, 05C20.

Key words and phrases. Parabolic structures, maximal flag manifolds, tournaments.
1
221
219
222
220
2 MARLIO PAREDES AND SOFÍA PINZÓN
A tournament T is cone–free if it does not contain any coned 3–cycle: we say

that a 3–cycle formed by the vertices i, j, k of T is coned if there is another vertex
x such that x beats i, j, k or loses to i, j, k.
Professor Brendan McKay pointed out to us that a tournament is cone–free if
and only if it is locally transitive (see [Br]). Then we have the following theorem,
equivalent to the previous result.
Theorem 1.2. The maximal flag manifold (F(n), J), n ≥ 4, admits an invari-
ant (1, 2)–symplectic metric if and only if the associated tournament T (J) is locally
transitive.
This theorem was proved in [CPP] and the proof is much simpler than the
proof for theorem 1.1. In [CNPPS], the concept of locally transitive tournament
was generalized to locally transitive digraphs and it was used to study f –structures
(see [Y]) on the classical flag manifold. In fact, the last theorem was generalized
to f –structures on a maximal flag manifold using the associated digraph to each
f –structure.
In this paper we prove that the associated tournament of a parabolic almost
complex structure on a maximal flag manifold has the following form

(n − k, . . . , n − k), if n = 2k − 1,
(n − (k + 1), . . . , n − (k + 1), n − k, . . . , n − k), if n = 2k.
This is a necessary condition for an almost complex structure to be parabolic.
Mo and Negreiros, in [MN], studied parabolic almost complex structures and
obtained the following result:
Theorem 1.3. Suppose that J is a parabolic invariant almost complex struc-
on F(n) = U (n)/(U (1) × · · · × U (1)). Then an invariant metric dsΛ =
2
ture
λ ω
i,j ij ij ⊗ ωij is (1, 2)–symplectic with respect to J if and only if Λ = (λ ij )
satisfies that
⎧
⎪ ai + ai+2 + · · · + ak−2 , if k − i ∈ 2N,
⎪
⎨ ak + ak+2 + · · · + an−1 + a1 + a3 + · · · + ai−2 , if i, n ∈ 2N − 1 and k ∈ 2N,
λik = ak + ak+2 + · · · + an + a2 + a4 + · · · + ai−2 , if n, k ∈ 2N − 1 and i ∈ 2N,
⎪
⎪
⎩ ak + ak+2 + · · · + an−1 + a1 + a3 + · · · + ai−2 , if n, k ∈ 2N and i ∈ 2N − 1,
ak + ak+2 + · · · + an−3 + an + a2 + a4 + · · · + ai−2 , if i, n ∈ 2N and k ∈ 2N − 1,
where a0 = an and a−1 = an−1 .
This theorem as well as its proof are not easy, the authors need to find condi-
tions on the matrix Λ = (λij ) in order to the almost complex structure admits a
(1, 2)–symplectic metric. In the present work, we prove directly that if the almost
complex structure is parabolic then the associated metric is (1, 2)–symplectic.
Also our results are strongly related with the important work “Invariant almost
Hermitian structures on flag manifolds”, due to San Martin and Negreiros [SMN],
in which, a complete classification of the invariant almost Hermitian structures on
flag manifolds is given. Another recent works related with this paper are [GM],
[N2] and [SN].
2. Tournaments
A tournament or n–tournament T , consists of a finite set T = {p1 , . . . , pn }
of n players or vertices together with a dominance relation, →, which assigns to
each pair of players a winner, that is, pi → pj or pj → pi . A tournament T can
TOURNAMENTS AND PARABOLIC ALMOST COMPLEX STRUCTURES. . . 221
223
3
• •
• •
• •
(0, 1)
(0, 1, 2) (1, 1, 1)
• • •
• •
• •
• • •
(0, 1, 2, 3) (1, 1, 1, 3) (0, 2, 2, 2) (1, 1, 2, 2)
Figure 1. Isomorphism classes of n–tournaments for n = 2, 3, 4.
be represented by a directed graph in which T is the set of vertices and any two
vertices are joined by an oriented edge. If the dominance relation is transitive, then
the tournament is called transitive. For a complete reference on tournaments see
the book by Moon [M].
Let T1 be a tournament with n players {1, . . . , n} and T2 be a tournament
with m players {1, . . . , m}. A homomorphism between T1 and T2 is a mapping
φ : {1, . . . , n} → {1, . . . , m} such that
T T
(2.1) 1
s −→ t =⇒ 2
(φ(s) −→ φ(t) or φ(s) = φ(t)).
When φ is bijective we say that T1 and T2 are isomorphic.

An n–tournament
n determines
a score vector (s1 , . . . , sn ), 0 ≤ s1 ≤ · · · ≤ sn ,
such that i=1 si = n2 , whose components represent the number of games won
by each player. Because of isomorphic tournaments have identical score vectors, we
can classify tournaments in isomorphism classes. Figure 1 shows the isomorphism
classes of n–tournaments for n = 2, 3, 4, together with their score vectors. In Figure
1 we use Moon’s notation in which not all of the arcs are included in the drawings.
If an arc joining two vertices has not been drawn, then it is to be understood that
the arc is oriented from the higher vertex to the lower vertex (see [M]).
Isomorphic tournaments have the same score vectors, but the reciprocal of this
statement is false because, for example, for n = 5 there are tournaments with the
same score vector which are not isomorphic, see Figure 2.
The canonical n–tournament Tn is defined by setting (see [BS])
i→j ⇐⇒ i < j.
It is easy to see that up to isomorphism, Tn is the only tournament to which the

following conditions are equivalent:
• the dominance relation is transitive, i.e. if i → j and j → k then i → k,
• there are no 3-cycles, i.e. close paths i1 → i2 → i3 → i1 ,
• the score vector is (0, 1, . . . , n − 1).
224
222
2 3 2 3 2 3
(1) (2) (3)
1 4 1 4 1 4
5 5 5
(0,1,2,3,4) (0,1,3,3,3) (0,2,2,3,3)
2 3 2 3 2 3
(4) (5) (6)
1 4 1 4 1 4
5 5 5
(0,2,2,2,4) (1,1,1,3,4) (1,1,2,2,4)
2 3 2 3 2 3
(7) (8) (9)
1 4 1 4 1 4
5 5 5
(1,1,2,3,3) (1,1,2,3,3) (1,2,2,2,3)
2 3 2 3 2 3
(10) (11) (12)
1 4 1 4 1 4
5 5 5
(1,2,2,2,3) (1,2,2,2,3) (2,2,2,2,2)
Figure 2. Isomorphism classes of 5–tournaments.
Given n ≥ 3 and a subset σ ⊂ {1, . . . , n}, the Grassmannian tournament

T (n, σ) is defined by
⎧
⎨ i < j for i ∈ σ, j ∈ σ c or i ∈ σ c , j ∈ σ,
(2.2) i → j ⇐⇒
⎩
i > j for i, j ∈ σ or i, j ∈ σ c .
These tournaments are used, in the following section, to define the parabolic
almost complex structures.
Given a tournament T , we say that a 3–cycle formed by the vertices i, j, k, of
T , is coned if there is another vertex l such that
(l → i, l → j and l → k) or (i → l, j → l and k → l).
We say that the tournament T is cone–free if and only if it does not contain any
coned 3–cycle.
Given a tournament T and a vertex i ∈ T we define the following subtourna-
ments
(2.3) T − (i) = {k ∈ T : k → i} and T + (i) = {k ∈ T : i → k},
which are called the in–neighbor and the out–neighbor of i respectively. T is called
locally transitive if and only if the subtournaments T − (i) and T + (i) are transitive
for each vertex i (see [Br]).
The following propositions are easy to prove, see [CPP].
225
5
Proposition 2.1. A tournament T is locally transitive if and only if it is

cone–free.
Proposition 2.2. A tournament T is is locally transitive if and only if all
4–subtournaments of T are locally transitive.
3. Flag manifolds
The classical maximal flag manifold is defined by
(3.1) F(n) = {(L1 , . . . , Ln ) : Li is a subspace of Cn , dimC Li = 1, Li ⊥ Lj }.
The unitary group U (n) acts transitively on F(n) turning this manifold into
the homogeneous space
U (n) U (n)
(3.2) F(n) = = ,
U (1) × U (1) × · · · × U (1) T
where T = U (1) × U (1) × · · · × U (1) is any maximal torus of U (n).
Let p be the tangent space of F(n) at the point (T ). It is known that u(n), the
Lie algebra of skew-hermitian matrices, decomposes as
u(n) = p ⊕ u(1) ⊕ · · · ⊕ u(1) ,
where p ⊂ u(n) is the subspace of zero-diagonal matrices.
In order to define any tensor on F(n) it is sufficient to give it on p, because
the action of U (n) on F(n) is transitive. An invariant almost complex structure on
F(n) is determined by a linear map J : p → p such that J 2 = −I and commutes
with the adjoint representation of the torus T on p.
For instance, in the case of
U (3) U (3)
F(3) = =
U (1) × U (1) × U (1) T
the tangent space is
⎧⎛ ⎞ ⎫
⎨ 0 z1 z2 ⎬
p = T (F(3))(T ) = ⎝ −z1 0 z3 ⎠ : z1 , z2 , z3 ∈ C ,
⎩ ⎭
−z2 −z3 0
then the application J : p → p given by
⎛ ⎞ ⎛ √ √ ⎞
0 z1 z2 √0 (− −1)z1 (−√ −1)z2
(3.3) ⎝ −z1 0 z3 ⎠ −→ ⎝ (−√−1)z1 ( −1)z3 ⎠
√ 0
−z2 −z3 0 (− −1)z2 ( −1)z3 0
is an almost complex structure on F(3).
Borel and Hirzebruch [BH] proved that the number of almost complex struc-
n
tures on F(n) is 2( 2 ) and this is the number of tournaments with n vertices. Burstall
and Salamon [BS] showed the relation between tournaments and almost complex
structures on F(n).
Given an invariant complex structure J, we define the associated tournament
T (J) in the following way: if J(aij ) = (aij ), then T (J) is such that for i < j
√ √
i → j ⇔ aij = −1 aij or i ← j ⇔ aij = − −1 aij ,
226
224
1 3
Figure 3. Associated tournament to almost complex structure (3.3).
see [MN]. For example, the tournament associated to the almost complex struc-
tures defined in (3.3) is the tournament in Figure 3. This beautiful result was
obtained by Burstall and Salamon in [BS]:
Theorem 3.1. An almost complex structures J on F(n) is integrable if and
only if the associated tournament is isomorphic to the canonical tournament.
Let us recall that an almost complex structure J is integrable if F(n) is a
complex manifold, i.e. F(n) admits complex coordinate systems with holomorphic
coordinate changes.
A well known result in the theory of tournaments says that a tournament is
isomorphic to canonical tournament if and only if it does not contain 3–cycles (see
[M]). Consequently, if T (J) contains a 3–cycle, then J is not integrable.
We consider Cn equipped with the standard Hermitian inner product, that is,

n
for V = (v1 , . . . , vn ) and W = (w1 , . . . , wn ) in Cn , we have V, W = vi wi . We
i=1
use the convention vı̄ = vi and fı̄j = fij̄ .
A frame consists of an ordered set of n vectors (Z1 , . . . , Zn ), such that Z1 ∧
. . . ∧ Zn = 0, and it is called unitary if Zi , Zj = δij̄ . The set of unitary frames
can be identified with the unitary group U (n).
If we write dZi = j ωij̄ Zj , the coefficients ωij̄ are the Maurer–Cartan forms
of the unitary group U (n). They are skew–Hermitian, that is, ωij̄ + ωj̄i = 0. For
more details see [ChW].
We may define all left–invariant metrics on (F(n), J) by (see [Bl] or [N1])

(3.4) ds2Λ = λij ωij̄ ⊗ ωı̄j ,
i,j
where Λ = (λij ) is a simetric real matrix such that

λij > 0, if i = j,
(3.5)
λij = 0, if i = j,
and the Maurer–Cartan forms ωij̄ are such that
T (J)
(3.6) ωij̄ ∈ C1,0 (forms of type (1,0)) ⇐⇒ i −→ j.
The metrics (3.4) are called Borel type and they are almost Hermitian for
every invariant almost complex structure J, this is, ds2Λ (JX, JY ) = ds2Λ (X, Y ) for
all tangent vectors X, Y . When J is integrable, ds2Λ is said to be Hermitian.
Let J be an invariant almost complex structure on F(n), T (J) the associated
tournament, and ds2Λ an invariant metric. The Kähler form with respect to J and
ds2Λ is defined by
(3.7) Ω(X, Y ) = ds2Λ (X, JY ),
227
7
for any tangent vectors X, Y . For each permutation τ of n elements, the Kähler
form can be written as follows (see [MN])
√
(3.8) Ω = −2 −1 µτ (i)τ (j) ωτ (i)τ (j) ∧ ωτ (i)τ (j) ,
i<j
where µτ (i)τ (j) = ετ (i)τ (j) λτ (i)τ (j) and

⎧
⎨ 1, if τ (i) → τ (j),
εij = −1, if τ (j) → τ (i),
⎩
0, if τ (i) = τ (j).
F(n) is said to be almost Kähler if and only if Ω is closed, that is, dΩ = 0. If
J is integrable and Ω is closed, then F(n) is said to be a Kähler manifold.
Mo and Negreiros proved in [MN] that

(3.9) dΩ = 4 Cτ (i)τ (j)τ (k) Ψτ (i)τ (j)τ (k) ,
i<j<k
where
(3.10) Cijk = µij − µik + µjk ,
and
(3.11) Ψijk = Im(ωij̄ ∧ ωı̄k ∧ ωj k̄ ).
p,q
We denote by C the space of forms of type (p, q) on F(n). Then, for any i, j, k,
we have either Ψijk ∈ C0,3 ⊕ C3,0 or Ψijk ∈ C1,2 ⊕ C2,1 . An invariant almost
Hermitian metric ds2Λ is said to be (1,2)–symplectic if and only if (dΩ)1,2 = 0. If
δΩ = 0, the codifferential of the Kähler form is zero, then the metric is said to be
cosymplectic.
In [CPP], was proved the following result
Theorem 3.2. Let (F(n), J, ds2Λ ) be the maximal flag manifold. The metric
ds2Λ is (1, 2)–symplectic if and only if the associated tournament T (J) is locally
transitive.
4. Parabolic almost complex structures

Burstall and Salamon, in [BS], define the parabolic almost complex structure
on F(n) as the correspondent structure to the Grassmannian tournament, defined
in (2.2), for σ = {1, 3, 5, . . .}. In [MN], Mo and Negreiros rewrote this definition
in the following way: an almost complex structure J on F(n) is called parabolic if
there is a permutation τ of n elements such that the associated tournament is, for
i < j, given by

τ (j) → τ (i), if j − i is even,
(4.1)
τ (i) → τ (j), if j − i is odd.
Figure 4 contains the corresponding tournaments to the parabolic almost com-
plex structures on F(3), F(4), F(5) and F(6), which have been constructed using
the last version of the definition.
The first result on parabolic almost complex structures is the following
Theorem 4.1. If J is a parabolic almost complex structure on F(n), n ≥ 4,
then T (J) contains k–subtournaments corresponding to parabolic structures for k =
3, 4, . . . , n − 1.
228
226
τ (2)
F (4)
F (3)
τ (3)
τ (2)
τ (1) τ (3) τ (1) τ (4)
(1,1,1) (1,1,2,2)
F (5) τ (2) τ (3) F (6) τ (2) τ (3)
τ (1)
τ (4)
τ (1) τ (4)
τ (5) τ (6) τ (5)

(2,2,2,2,2) (2,2,2,3,3,3)
Figure 4. Parabolic tournaments for n = 3, 4, 5, 6.
Proof. We use induction on n, beginning with n = 4. Clearly, the parabolic

4–tournament contains a parabolic 3–tournament. Suppose that the result is true
for n then, by the induction hypotheses, the parabolic n tournament contains pa-
rabolic k–subtournaments for k = 3, 4, . . . , n − 1. So, we only need to show that
the parabolic (n + 1)–tournament contains a parabolic n–tournament. Without
loss of generality, we can suppose that the vertices are enumerated with numbers
1, 2, . . . , n, n + 1 in such a way that every subtournament with vertices 1, 2, . . . , k,
which we denote by (12 . . . k), is parabolic for k = 1, 2, . . . , n − 1. Now we go to
prove that the subtournament (12 . . . n) is parabolic. Suppose that the tournament
(12 . . . n) is not parabolic, then we have two possibilities:
(a) there exists i ∈ {1, 2, . . . , n − 1} such that i → n and n − i is even, but it
is not possible because the (n + 1)–tournament is parabolic and n → i.
(b) there exists i ∈ {1, 2, . . . , n − 1} such that n → i and n − i is odd, but it
is not possible because the (n + 1)–tournament is parabolic and i → n.

The parabolic 3–tournament is a 3–cycle, see Figure 4, therefore the previous

theorem implies that every parabolic tournament contains a 3–cycle. Then we have
the following result
Corollary 4.2. All of parabolic almost complex structures on F(n) are not
integrable.
The following theorem gives us a necessary condition to an almost complex

structures to be parabolic as pointed out in the introduction.
229
9
Theorem 4.3. If J is a parabolic almost complex structure on F(n) then the

score vector of the associated tournament T (J) has the following form

(n − k, . . . , n − k), if n = 2k − 1,
(4.2)
(n − (k + 1), . . . , n − (k + 1), n − k, . . . , n − k), if n = 2k.
Proof. In order to facilitate the calculations we consider the order of the
vertices of T (J), then the score vector take the following form

(n − k, . . . , n − k), if n = 2k − 1,
(n − k, n − (k + 1), . . . , n − k, n − (k + 1)), if n = 2k.
As in the last theorem, we use induction over n again. According to Figure 4, the
result is true for n = 3, 4, 5. Suppose that the result is true for n, then we have two
cases:
(a) If n = 2k − 1, then the score vector of T (J) is (n − k, . . . , n − k). As
n + 1 = 2k, by the definition of parabolic almost complex structure we
have
(n + 1) − n = 1 =⇒ n → n + 1,
(n + 1) − (n − 1) = 2 =⇒ n + 1 → n − 1,
..
.
(n + 1) − 2 = n − 1 = 2k − 2 =⇒ n + 1 → 2,
(n + 1) − 1 = n = 2k − 1 =⇒ 1 → n + 1.
This implies that n + 1 wins k − 1 = n − k games and loses k games. Also,
we have that n + 1 wins to i if i is even and loses with i if i is odd. Then
the score vector for n + 1 is
(n − k + 1, n − k, . . . , n − k + 1, n − k) =
= ((n + 1) − k, (n + 1) − (k + 1), . . . , (n + 1) − k, (n + 1) − (k + 1)).
(b) If n = 2k, then the score vector of T (J) is
(n − k, n − (k + 1), . . . , n − k, n − (k + 1)).
Now we use a similar argument to the previous case; as n + 1 = 2k + 1 =
2(k + 1) − 1 then, by the definition of parabolic structure we have
(n + 1) − n = 1 =⇒ n → n + 1,
(n + 1) − (n − 1) = 2 =⇒ n + 1 → n − 1,
..
.
(n + 1) − 2 = n − 1 = 2k − 1 =⇒ 2 → n + 1,
(n + 1) − 1 = n = 2k =⇒ n + 1 → 1.
In this case n + 1 beats k = n − k games and loses to k games. Also, we
have that n + 1 beats i, if i is odd, and loses to i, if i is even. Then the
score vector for n + 1 is
(n−k, n−(k+1)+1, . . . , n−k, n−(k+1)+1, n−k) = (n−k, . . . , n−k).

230
228
(2, 2, 2, 3, 3, 3)
Figure 5. Tournament with the same score vector of the parabolic tournament.
The reciprocal of this theorem is false. For example, for n = 6, there are
tournaments different to the parabolic with the same score vector. The tournament
in Figure 5 is an example of a 6–tournament with the same score vector of the
parabolic tournament which is not isomorphic to it.
Finally, we show that all of the parabolic almost complex structures admit
(1, 2)–symplectic metrics. In order to prove this, we will show that the associated
tournament T (J) to the almost complex structure J is locally transitive.
Theorem 4.4. If J is a parabolic almost complex structure on F(n), then the

metric ds2Λ is (1, 2)–symplectic.
Proof. Using Theorem 3.2 we need to prove that the associated tournament
T (J) is locally transitive. Suppose that T (J) is not locally transitive, then there
is a vertex i such that T (J)+ (i) is not transitive or T (J)− (i) is not transitive. If
T (J)+ (i) is not transitive then it contains a 3–cycle x → y → z → x and i → x,
i → y, i → z. By the definition of parabolic almost complex structure we have the
following options:
(1) i < x < y < z, then x − i, y − i and z − i are odd. Hence, (y − i) − (x − i) =
y −x is even and the definition of the parabolic tournament implies y → x,
which is a contradiction.
(2) x < i < y < z, then i − x is even and y − i, z − i are odd. Hence,
(z − i) − (y − i) = z − y is even and the definition of the parabolic
tournament implies z → y, which is a contradiction.
(3) x < y < i < z, then i − x, i − y are even and z − i is odd. Hence,
(i − x) − (i − y) = y − x is even and the definition of the parabolic
tournament implies y → x, which is a contradiction.
(4) x < y < z < i, then i−x, i−y and i−z are even. Hence, (i−x)−(i−y) =
y −x is even and the definition of the parabolic tournament implies y → x,
which is a contradiction.
In case T (J)− (i) is not transitive, we can use a similar argument.
Acknowledgments. We would like to thank Professor David Colón

Arroyo for his help reviewing the English language in this work.
231
11
References
[Bo] A. Borel, Kählerian coset spaces of semi–simple Lie groups, Proc. Nat. Acad. of Sci. USA
40 (1954), 1147–1151.
[BH] A. Borel and H. Hirzebruch, Characteristic classes and homogeneous spaces I, Amer. J.
Math. 80 (1958), 458–538.
[Bl] M. Black, Harmonic maps into homogeneous spaces, Pitman Res. Notes Math. Ser., vol. 255,
Longman, Harlow, 1991.
[Br] A. E. Brouwer, The enumeration of locally transitive tournaments, Afdeling Zuivere
Wiskunde [Department of Pure Mathematics], 138 (1980), Mathematisch Centrum, Ams-
terdam.
[BS] F.E. Burstall and S. Salamon, Tournaments, flags and harmonic maps, Math. Ann. 277
(1987), 249–265.
[ChW] S.S. Chern and J.G. Wolfson, Harmonic maps of the two–sphere into a complex Grass-
mann manifold II, Ann. of Math. 125 (1987), 301–335.
[CNPPS] N. Cohen, C.J.C. Negreiros, M. Paredes, S. Pinzón and L.A.B. San Martin, f –Structures
on the classical flag manifold which admit (1, 2)–symplectic metrics, Tohoku Math. J. (2) 57,
No. 2 (2005), 261–271.
[CNS] N. Cohen, C.J.C. Negreiros, and L.A.B. San Martin, (1, 2)–symplectic metrics, flag mani-
folds and tournaments, Bull. London Math. Soc. 34 (2002), 641–649.
[CPP] N. Cohen, M. Paredes and S. Pinzón, Locally transitive tournaments and the classification
of (1, 2)–symplectic metrics on maximal flag manifolds, Illinois J. Math. 48, No. 4 (2004),
1405–1415.
[GH] A. Gray and L. M. Hervella, The sixteen classes of almost Hermitian manifolds and their
linear invariants, Ann. Mat. Pura Appl. (4) 123 (1980), 35–58.
[GM] L. Grama and R. M. Martins, The Ricci flow of left–invariant metrics on full flag manifold
SU (3)/T from a dynamical systems point of view, Bull. Sci. Math. 133 (2009), 463–469.
[M] J.W. Moon, Topics on tournaments, Holt, Rinehart and Winston, New York, 1968.
[MN] X. Mo and C.J.C. Negreiros, (1, 2)-Symplectic structures on flag manifolds, Tohoku Math.
J. (2), 52 (2000), 271–282.
[N1] C.J.C. Negreiros, Some remarks about harmonic maps into flag manifolds, Indiana Univ.
Math. J. 37 (1988), No. 3, 617–636.
[N2] C.J.C. Negreiros, Stability of holomorphic–horizontal maps and Einstein metrics on flag
manifolds, Rev. Un. Mat. Argentina 47 (2006), No. 2, 85–94.
[P1] M. Paredes, Torneos y estructuras parabólicas sobre variedades bandera maximales, Rev.
Integr. Temas Mat. 17 (1999), 1–10.
[P2] M. Paredes, Aspectos da geometria complexa das variedades bandeira, Doctoral Thesis, State
University of Campinas, Brazil, 2000.
[P3] M. Paredes, Some results on the geometry of full flag manifolds and harmonic maps, Rev.
Col. Mat. 34 (2000), 57–89.
[P4] M. Paredes, Families of (1, 2)–symplectic metrics on full flag manifolds, Internat. J. Math.
Math. Sci. 29 (2002), 651–664.
[Pi] S. Pinzón, Integrability of f–structures on generalized flag manifolds, Rev. Un. Mat. Argentina
47 (2006), 99–113.
[SMN] L.A.B. San Martin and C.J.C. Negreiros, Invariant almost Hermitian structures on flag
manifolds, Adv. Math. 178 (2003), 277–310.
[SN] E.C.F. dos Santos and C.J.C. Negreiros, Einstein metrics on flag manifolds, Rev. Un. Mat.
Argentina 47 (2006), No. 2, 77–84.
[Y] K. Yano, On a structure defined by a tensor field of type (1, 1) satisfying F 3 + F = 0, Tensor
14 (1963), 99–109.
Department of Mathematics, Universidad del Turabo, PO Box 3030, Gurabo, Puerto

Rico 00778-3030, USA
E-mail address: maparedes@suagm.edu
Escuela de Matemáticas, Universidad Industrial de Santander, A.A. 678, Bucara-

manga, Colombia
E-mail address: spinzon@uis.edu.co
This volume represents the 2007–2008 Jairo Charris Seminar in Algebra and Analysis on
Differential Algebra, Complex Analysis and Orthogonal Polynomials, which was held at
the Universidad Sergio Arboleda in Bogotá, Colombia.
It provides the state of the art in the theory of Integrable Dynamical Systems based on
such approaches as Differential Galois Theory and Lie Groups as well as some recent
developments in the theory of multivariable and q -orthogonal polynomials, weak Hilbert’s
16th Problem, Singularity Theory, Tournaments in flag manifolds, and spaces of bounded
analytic functions on the unit circle.
The reader will also find survey presentations, an account of recent developments, and the
exposition of new trends in the areas of Differential Galois Theory, Integrable Dynamical
Systems, Orthogonal Polynomials and Special Functions, and Bloch–Bergman classes of
analytic functions from a theoretical and an applied perspective.
The contributions present new results and methods, as well as applications and open prob-
lems, to foster interest in research in these areas.
IMA on the Web

http://ima.usergioarboleda.edu.co
CONM/509 AMS on the Web

www.ams.org

Ontemporary Athematics: Differential Algebra, Complex Analysis and Orthogonal Polynomials

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ontemporary Athematics: Differential Algebra, Complex Analysis and Orthogonal Polynomials

Uploaded by

Copyright:

Available Formats

CONTEMPORARY

American Mathematical Society

American Mathematical Society

American Mathematical Society

American Mathematical Society

Library of Congress Cataloging-in-Publication Data

Primitivo B. Acosta-Humánez, IMA - Universidad Sergio Arboleda

Diﬀerential Galois Theory of Algebraic Lie-Vessiot Systems

David Blázquez-Sanz and Juan José Morales-Ruiz

This paper is dedicated to Jairo Antonio Charris.

Abstract. In this paper we develop a diﬀerential Galois theory for algebraic

where A1 , . . . , As are autonomous vector ﬁelds which inﬁnitesimally span a pre-

interpretation of diﬀerential equations as schemes with derivations, and their solu-

Notation and Conventions. We denote diﬀerential and ordinary ﬁelds and

2. Algebraic Groups and Homogeneous Spaces

2.3. Algebraic Homogeneous spaces.

2.4. Existence of quotients: Chevalley’s theorem. Let V ve a C-vector

such that H is the stabilizer of T ,

2.5. Galois Cohomology. In this section, we assume that C is a perfect ﬁeld ;

is a C-point of the homogeneous space G/G . This x is the class of a unique C- ¯

Definition 2.8. We call algebra of fundamental ﬁeld R(G, M ) to the Lie

There is a canonical surjective Lie algebra morphism,

3. Diﬀerential Algebraic Geometry

Theorem 3.3. Let A be a diﬀerential ring. The following are equivalent:

Then, λi ∈ AU and K[CAU ] = K[λ1 , . . . , λs ]. We will prove that CAU = C[λ1 , . . . , λs ].

Proposition 3.11. Assume A is Keigher. If X  is connected, then X is con-

for any open subset U ⊆ X.

3.11. Galois Correspondence for Strongly Normal Extensions. Let us

Definition 3.27. A scheme with derivation is a pair (X, ∂X ) consisting of a

denote U  the diﬀerential spectrum DiﬀSpec(OX (U )) that we consider as a subset

Proposition 3.33. Suppose (X, ∂) is Keigher, then

Definition 3.34. (X, ∂) is almost-constant if Const(X, ∂) is a scheme.

4. Galois theory of Algebraic Lie-Vessiot Systems

4.1. Diﬀerential Algebraic Dynamical Systems. Here we establish a par-

Definition 4.1. A diﬀerential algebraic dynamical system is a K-scheme with

for certain elements fi ∈ K and X

Definition 4.5. A non-autonomous algebraic vector ﬁeld X in M with coef-

Definition 4.6. We call automorphic vector ﬁelds to the Lie-Vessiot vector

From now on let X be a Lie-Vessiot vector ﬁeld in M , with coeﬃcients in K,

4.3. Logarithmic Derivative. A K-point of the algebraic group G has co-

Let us consider σ ∈ G(K) and the canonical morphism σ of taking values in

Lemma 4.8. The commutator σ  = [∂, σ ] of the diagram (4.1) is a derivation

If σ is a geometric point of GK , then Rσ−1 is a automorphism of GK sending σ

Definition 4.9. Let σ be a geometric point of GK ; we call logarithmic de-

(2) Let us consider σ and τ in G(K):

4.4. Automorphic Equation.

OGL ,σ . The composition with σ give us the tangent vector B

Furthermore, B is determined by the commutator B σ = [∂, σ ] and then it is unique

for the so-called

4.5. Solving Lie-Vessiot Systems.

Lemma 4.15. Let Z be a C-algebraic variety and (ZK , D) a non-autonomous

splits. It implies that there exist an C-scheme

Lemma 4.16. Let Z be a reduced C-scheme. There is a one-to-one correspon-

Lemma 4.17. Let Z be a C-algebraic variety and (ZK , D) a non autonomous

Proof. By substituting Z for certain open subset we can assume that Y is

Theorem 4.19. Assume that the action of G on M is faithful. Then the

4.6. Splitting Field of an Automorphic System. Note that a diﬀerential

4.7. Action of G(C) on GK . For each σ ∈ G(C), Rσ is an automorphism of

Proof. Let x be a closed point. Then the canonical morphism x of taking

The morphism σ is the dual of a morphisms σ from Spec(κ(σ)) to U . In other

and thus, a commutative diagram of diﬀerential schemes,

od Diﬀ(GL(n, K), ∂A ) is the set of maximal diﬀerential ideals of R. A Picard-Vessiot

Lemma 4.31. Let K ⊂ S be a splitting extension. The canonical projection

is a C-point of the homogeneous space G/G . This x is the class of a unique C- ¯

Proposition 3.11. Assume A is Keigher. If X is connected, then X is con-

denote U the diﬀerential spectrum DiﬀSpec(OX (U )) that we consider as a subset

Lemma 4.8. The commutator σ = [∂, σ ] of the diagram (4.1) is a derivation

od Diﬀ(GL(n, K), ∂A ) is the set of maximal diﬀerential ideals of R. A Picard-Vessiot

π : Diﬀ(GS , ∂A ) → Diﬀ(GK , ∂A )

π : Diﬀ(GF , ∂A ) → Diﬀ(GK , ∂A ),

Diﬀ(GK , ∂A ) ×C G → Diﬀ(GK , ∂A ).

Definition 4.34. Let x ∈ Diﬀ(GK , ∂A ) be a Kolchin closed diﬀerential point.

Proposition 4.35. Galx (GK , ∂A ) is an algebraic subgroup of G.

Theorem 5.5. Let us consider an algebraic subgroup G of G verifying:

belongs to R(Gal0x (GK , ∂A )) ⊗C K◦ .

Gal(L/M) Galx (GK , ∂A ) ∩ N.

Gr(E, m) Gr(E ∗ , n − m).

Rϕ(x)·R /ϕ(x) · Rϕ(x)·R = (R /ϕ(x) · R )ϕ(x)·R = R ⊗R κ(x).